close

Вход

Забыли?

вход по аккаунту

?

Local methods for the cosmic microwave background

код для вставкиСкачать
Motion Optimized Conformal Microwave Imaging for Biomedical
Applications
A Thesis
Submitted to the Faculty
in partial fulfillment of the requirements for the
degree of
Doctor of Philosophy
by
NEIL R. EPSTEIN
Thayer School of Engineering
Dartmouth College
Hanover, New Hampshire
MAY 2013
Examining Committee:
Chairman_______________________
Paul M. Meaney
Member________________________
Keith D. Paulsen
Member________________________
Alex Hartov
Member________________________
Mikael Persson
___________________
F. Jon Kull
Dean of Graduate Studies
UMI Number: 3613688
All rights reserved
INFORMATION TO ALL USERS
The quality of this reproduction is dependent upon the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
UMI 3613688
Published by ProQuest LLC (2014). Copyright in the Dissertation held by the Author.
Microform Edition © ProQuest LLC.
All rights reserved. This work is protected against
unauthorized copying under Title 17, United States Code
ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
Abstract
Investigations into alternative breast cancer (BC) imaging techniques have
become increasingly popular based on the limitations of traditional imaging modalities:
X-ray mammography uses ionizing radiation, has limited intrinsic contrast and is
associated with high false-positive and false-negative rates. Microwave tomographic
imaging (MTI) has the ability to detect a wide range of dielectric property (DP) values,
and due to the contrast that exists between the DPs of normal and abnormal breast tissue,
MTI has shown promise as an alternative BC imaging modality. This thesis reports on a
third generation system currently used in clinical trials at Dartmouth Hitchcock Medical
Center. This system’s improvements include increased data acquisition speeds and
capabilities due to upgraded microwave electronic components and motion control
hardware, respectively. Part I of this work will evaluate the system’s microwave
electronics in terms of channel isolation, system sensitivity and measurement
repeatability in an effort to define an optimal operational bandwidth. The system’s
antenna array is composed of two interwoven sub-arrays (SAs) that can independently
move to a number of positions. We have found that incorporating measurement data
obtained at larger SA separation spacing during 3D acquisition results in unwanted
artifacts in the reconstructed images. S-parameter studies have indicated that signals
transmitted at these larger separation distances fall below the noise floor (NF) of the
system’s receiving channels. Part II of this thesis will focus on analyzing measurement
sensitivity as a function of increasing SA spacing. The analysis has resulted in the
creation of a motion optimized imaging system; increasing examination speeds by
eliminating measurements positions where signals fall below the NF. Additionally, we
ii
have seen that our reconstruction algorithm has benefited from the incorporation of
boundary information regarding the object under test (OUT). An optical-scanning system
that can capture the boundary of the OUT while submerged in the system imaging
chamber has been developed and mounted to the new imaging prototype. Conforming the
reconstruction property mesh to the boundary of the OUT has increased the accuracy of
recovered DPs. Part III of this thesis evaluates the boundary conformed microwave
reconstruction process utilizing information obtained from the integrated optical-scanner.
iii
Acknowledgements
This work would not have been possible without the guidance provided by my
advisors’ Dr. Paul Meaney and Dr. Keith Paulsen. Their knowledge regarding the
computational and hardware aspects associated with microwave imaging (MI) are of the
highest caliber, and my success is a function of their experience and expertise. I greatly
appreciate all the input and recommendations they have provided during my time at
Dartmouth.
Working with current MIS group member Matt Pallone has been an incredibly
rewarding experience. He is an elite research engineer; collaborating has been a privilege.
His hardware and software skills were essential in developing the MI system presented in
this thesis; I am truly proud of the work we successfully conducted together. Peter Jensen
from the Technical University of Denmark has positively impacted this project and I
would like to thank him for his input and insightful discussions regarding MT imaging.
Additional I would like to thank and recognize Dr. Matthew McGarry for his constant
feedback and willingness to discuss all relevant topics throughout our Ph.D. careers. His
computational knowledge is unmatched and I truly value his input and friendship.
Most importantly, I am grateful and indebted to my parents Lawrence and Debra
Epstein, my grandparents Melvin and Pheobe Rabinowitz and my girlfriend Dr. Marcella
Lucas; their guidance and support regarding all aspects of life during my graduate career
were unmatched. My motivation to succeed is founded on their unyielding inspiration
and influence. I feel privileged to have had their encouragement and involvement in this
process; I hope all students have such support from their loved ones. Among an infinitely
long list of supporters and support roles, I am specifically thankful for my Father’s help
iv
in editing this thesis and my girlfriend’s support while writing it. Lawrence deserves an
honorary doctorate for the information he has learned and the input he provided during
this process. Marcella’s impact on my life is second to none; truly inspirational. She
constantly challenges me and has earned my upmost respect.
v
Table of Contents
1. INTRODUCTION......................................................................................................... 1
1.1 MOTIVATION FOR DEVELOPING ALTERNATIVE BREAST CANCER IMAGING MODALITIES
......................................................................................................................................... 6
1.2 PHYSICAL BASIS UNDERLYING THE USE OF ELECTROMAGNETIC ENERGY FOR
DIELECTRIC PROPERTY ASSESSMENT ................................................................................ 9
1.3 BREAST TISSUE DIELECTRIC PROPERTIES .................................................................. 16
1.4 DIELECTRIC CONTRAST IMAGING USING MICROWAVE ENERGY................................. 19
1.4.1. Ultra-wide band and space-time imaging techniques .................................... 22
1.4.2. Tomographic imaging techniques ................................................................... 28
1.5 SUMMARY OF PROPOSED WORK................................................................................ 33
2. MICROWAVE TOMOGRAPHIC IMAGING: A SYSTEM OVERVIEW ......... 35
2.1 SYSTEM HARDWARE LAYOUT ................................................................................... 38
2.2 ELECTRONIC HARDWARE DESIGN ............................................................................. 41
2.2.1. Microwave source and LO network ................................................................ 43
2.2.2. Transmitting/receiving channels..................................................................... 46
2.2.3. System integration ........................................................................................... 49
2.3 ILLUMINATION CHAMBER AND LIQUID-COUPLED CLINICAL INTERFACE .................... 51
2.3.1. Antenna design and characterization ............................................................. 55
2.4 COMPUTER CONTROL AND DATA EXTRACTION ......................................................... 58
2.5 HOMOGENEOUS DATA CALIBRATION ........................................................................ 60
2.6 IMAGE RECONSTRUCTION ........................................................................................ 61
2.7 CONCLUSION ............................................................................................................ 64
vi
3. ELECTRONIC HARDWARE PERFORMANCE .................................................. 65
3.1 SYSTEM SENSITIVITY AND ERROR ............................................................................. 66
3.2 INTER-CHANNEL ISOLATION ..................................................................................... 68
3.3 MEASUREMENT REPEATABILITY AND EFFECTIVE OPERATING BANDWIDTH .............. 84
3.4 ADDITIONAL ISOLATION AND LEAKAGE PATH ANALYSIS .......................................... 85
3.5 CONCLUSION ......................................................................................................... 89
4. MOTION CONTROL SYSTEM AND PERFORMANCE .................................... 90
4.1 INDEPENDENT SUB-ARRAY CONTROL ....................................................................... 92
4.2 2D MOVEMENT STRATEGY ....................................................................................... 95
4.3 3D MOVEMENT STRATEGY ....................................................................................... 98
4.4 SUB-ARRAY PLANE-SEPARATION DISTANCE EVALUATION ...................................... 104
4.5 3D IF SIGNAL AND SUB ARRAY PLANE SEPARATION DISTANCE ANALYSIS .............. 110
4.6 CONCLUSION .......................................................................................................... 119
5. MOTION CONTROL SYSTEM IMPROVEMENTS .......................................... 120
5.1 UPDATED MOTION CONTROL SYSTEM ................................................................... 122
5.2 RS232 COMMUNICATION CONTROLLED WITH CME2 SOFTWARE ........................... 125
5.3 RS232 COMMUNICATION CONTROL WITH LABVIEW SOFTWARE ........................... 127
5.4 CAN ADAPTER COMMUNICATION CONTROLLED WITH LABVIEW SOFTWARE........ 129
5.5 CONCLUSION .......................................................................................................... 131
6. INTEGRATING UPDATED MOTION CONTROL SYSTEM AND MOUNTED
OPTICAL SCANNER .................................................................................................. 132
6.1 MASTER AND SLAVE COMPUTER SYSTEM COMMUNICATION ................................. 136
vii
6.2 MASTER COMPUTER CONTROL AND OPERATION ................................................... 138
6.3 SLAVE COMPUTER CONTROL AND OPERATION ...................................................... 140
6.4 CONCLUSION .......................................................................................................... 143
7. CONFORMAL MICROWAVE IMAGING .......................................................... 144
7.1. 2D SIMULATION STUDIES...................................................................................... 146
Case 1: A single inclusion....................................................................................... 148
Case 2: Two inclusions ........................................................................................... 150
Case 3: Three inclusions......................................................................................... 153
7.2. PHANTOM STUDIES ............................................................................................... 156
Parameter mesh scaling and translation ................................................................ 159
7.3. 3D SIMULATION STUDIES...................................................................................... 163
7.4. SURFACE MOUNTED LASER SCANNING SYSTEM STUDIES ..................................... 165
7.5 CONCLUSION .......................................................................................................... 169
8. SUMMARY ............................................................................................................... 170
REFERENCES.............................................................................................................. 171
viii
List of Figures
Figure 1.1. Dispersion relations of and
for typical a biological tissue (compliments of
science.direct.com). .......................................................................................................... 11
Figure 1.2. Summary of measured DPs for normal (black) and malignant (red) breast
tissue at radio and microwave frequencies. ...................................................................... 15
Figure 1.3. Bar charts of the mean imaged DP property contrast ratios using Dartmouth’s
MI system. Vertical lines represent the 95% confidence interval, and mean numerical
value inside each box. P values appear above the bars; cancers have the highest mean
contrast ration relative to benign breast abnormalities and normal control subjects. ....... 15
Figure 1.4. Plot of power absorption ratio of malignant to normal tissue for various tissue
types. ................................................................................................................................. 18
Figure 1.5. (a) University of Bristol’s UWB imaging system highlighting their
backscatter interface and symmetrically curved antenna array and (b) patient lying prone
with breast inserted in imaging system. ............................................................................ 21
Figure 1.6. (a) X-ray mammogram highlighting abnormal region and (b) backscatter
intensity map showing abnormal region in red. ................................................................ 22
Figure 1.7. University of Bristol’s new experimental UWB imaging system highlighting
the VNA, switching-bank and modified patient interface. ............................................... 23
Figure 1.8. University of Calgary’s TSAR system used to scan volunteers. .................... 24
ix
Figure 1.9. (a) Top and (b) side views of Calgary’s TSAR prototype system with
dimension; additional antenna locations are shown (shaded) to illustrate the tank
rotation…………………………………………………………………………………...25
Figure 1.10. Photograph of the experimental illumination chamber and antenna array
configuration. .................................................................................................................... 28
Figure 1.11. (a) Photograph of Technical Universty of Denmark’s MI system and (b)
photograph of the measurement unit with shielding fences removed............................... 29
Figure 2.1. Photographs and computer renderings of Dartmouth’s MT imaging system
highlighting (a) 16-element monopole antenna housed in associated illumination
chamber, (b) schematic of the two interwoven SAs, (c) SA attachment to independent
linear actuator-controlled mounting-plates, (d) SAs in an in-plane configuration and (e)
SAs in an out-of-plane configuration. ............................................................................... 37
Figure 2.2. System connectivity diagram highlighting the RF, LO and antenna inputs to
the transceiver units, as well as the system’s parallelization. ........................................... 38
Figure 2.3. (a) Photograph of Dartmouth’s MT system highlighting its modular design
including 1-illumination chamber, 2-microwave electronics’ compartment, 3-DAQ
components, 4-power supply arrangement and (b) photograph of the system covered with
customized patient table with breast aperture. .................................................................. 40
Figure 2.4. Diagram of new LO PD network highlighting the streamlined design
configuration. .................................................................................................................... 43
Figure 2.5. CAD drawings of the (a) 2-way and (b) 8-way PDs (compliments of Pulsar
Microwave Corporation). .................................................................................................. 44
x
Figure 2.6. Diagram of the new RF switching matrix highlighting the bi-level SP4T
switch configuration and (b) a photograph of a single SP4T switch. ............................... 45
Figure 2.7. A schematic of the system’s transceiver unit diagraming the layout of its
electronic components, including amplifiers, switches, and a mixer; also shown is the RF,
LO, Antenna Port and A/D board connection points. ....................................................... 46
Figure 2.8. Photographs of (a) mounted modular assembly of 1- RF switch matrix
module, 2- LO PD module and 3- Wall mounted transceiver unit assembly, (b) Bulkhead
mounting plate used for transceiver-antenna connection as seen from the microwave
electronics’ compartment, (c) Bulkhead mounting plate used for transceiver-antenna
connection as seen from the illumination chamber cabinet, and (d) Mounting module
terminal block connection. ................................................................................................ 48
Figure 2.9. Photographs of (a) commercial electronic hardware cabinet including 1- DAQ
chassis, 2- DC power supplies, 3- RF signal generator and (b) positioning of microwave
and non-microwave electronics’ cabinets. ........................................................................ 49
Figure 2.10. Photograph of the 3rd generation system’s clinical interface including 1illumination chamber, 2- antenna array and 3- coupling medium. ................................... 50
Figure 2.11. Photograph of the antennas (a) entering the illumination tank from the
bottom of the imaging apparatus through the base plate holes and seal-related O-rings and
(b) a top-down view of the antenna array inside the tank with Teflon seals in
place……………………………………………………………………………………...51
Figure 2.12. Photographs of (a) the pump used for filling and draining the tank and (b)
the valve connection from the coupling mediums storage container to the imaging tank..
........................................................................................................................................... 53
xi
Figure 2.13. (a) CAD drawing of our system’s monopole antenna (compliments of Dr.
Tain Zhou) and (b) photograph the two monopole antennas utilized in our dual SA design
(also shown is one monopole with its active part exposed). ............................................. 54
Figure 2.14. Photograph of the system’s DAQ chassis showing Ethernet connection,
digital I/O connection and IF signal inputs to the two 8-channel 24-bit dynamic signal
acquisition boards. ............................................................................................................ 57
Figure 3.1. Schematic of the imaging system highlighting the incorporation of the new
RF and LO networks (only shown for four transceiver connections). .............................. 63
Figure 3.2. System (a) amplitude sensitivity response and (b) phase error as a function of
variable input power levels. .............................................................................................. 65
Figure 3.3. Schematic of the new bi-level RF switching matrix highlighting its two level
design and the connection of channels to their associated SP4T switch (shown for
channels 1 to 4). ................................................................................................................ 66
Figure 3.4. Plot of measured signal levels (normalized to the desired signal) with channel
#1 directly connected to channel #13 (relative receiver #12) and all other channels
terminated with 50
loads. .............................................................................................. 67
Figure 3.5. Schematic of the experimental set-up used to measure the IF signal output
from transceiver unit #2 (LO connections are not displayed and antenna ports are shown
terminated with matched loads). ....................................................................................... 68
Figure 3.6. Photograph of the experimental set-up shown in figure 3.5 highlighting the
detected leakage signal at transceiver unit #2’s IF port. ................................................... 69
xii
Figure 3.7. Schematic of the modified experimental set-up used to measure the IF signal
output from transceiver unit #2 (LO connections are not displayed and antenna ports are
shown terminated with matched loads)............................................................................. 69
Figure 3.8. Photograph of modified experimental set-up shown in figure 3.7 highlighting
the modified signal output at transceiver unit #2’s IF port. .............................................. 70
Figure 3.9. Schematic of the experimental set-up used to measure the isolation between
the active and non-active channels of a single active SP4T switch (shown connected to
channels #1 and #2). Isolation measurements for Channels #3 and #4 were measured
using the same set-up (by replacing #2 with #3 and #4, respectively). ............................ 71
Figure 3.10. S21 measurements of each non-active channel of a transmitting SP4T switch;
each channel has greater than 40 dB of isolation. ............................................................. 71
Figure 3.11. Schematic of the experimental set-up used to measure the isolation between
the active channel of a 1st level SP4T switch and the non-active channels of an active 2nd
level SP4T switch (shown connected to Channels #1 and #3). Isolation measurements for
Channels #2 and #4 were measured using the same set-up (by replacing #3 with #2 and
#4, respectively). ............................................................................................................... 72
Figure 3.12. S21 measurements taken between the transmitting channel of the 1st level
SP4T switch and the non-transmitting channels of the active 2nd level SP4T switch. ..... 73
Figure 3.13. Schematic of the transceiver unit highlighting the incorporation of the SPST
switch used to increase the isolation between its transmitting and receiving pathways. .. 74
Figure 3.14. Schematic of the system’s circular antenna configuration highlighting the
transmitter-receiver distance variations (Compliments of Dr. Fox). ................................ 75
xiii
Figure 3.15. Detected leakage signal through the RF switching matrix at channels (a) two,
(b) three, (c) four, (d) five, (e) six and (f) thirteen. ........................................................... 76
Figure 3.16. Overlay of DAQ measured transmission and leakage data for each channel
of the parallel detection scheme. ....................................................................................... 77
Figure 3.17. (a) SNR for each channel of the parallel detection scheme as a function of
frequency........................................................................................................................... 78
Figure 3.18. Schematic of the modified experimental set-up incorporating an additional
40 dB of attenuation on each non-transmitting channel of the active switch (shown in
black)................................................................................................................................. 80
Figure 3.19. Plot of measured signals with channel #1 directly connected to channel #13
(relative receiver #12), and all other channels terminated with 50
loads...................... 81
Figure 3.20. Photograph of the breast phantom submerged in the imaging system’s
illumination chamber. ....................................................................................................... 82
Figure 3.21. Schematic of two possible leakage paths through two adjacent transceiver
units: Antenna signals received by transceiver A leaked into the IF output of transceiver
B (green and red lines), and RF input to transceiver A leaked to IF output of transceiver B
(blue and red lines)............................................................................................................ 84
Figure 3.22. S21 measurements of a transceiver unit taken between its antenna and LO
ports (shown in blue), and RF and LO ports (shown in red). ........................................... 86
Figure 4.1. Photographs of the imaging construct highlighting its (a) independent SA
mounting plates and antenna attachments, connected to the motion control system’s linear
actuators and (b) low–profile design positioned inside the patient housing unit. ............. 90
xiv
Figure 4.2. Photograph of (a) phantom set-up and (b) phantom submerged in the imaging
system’s illumination chamber. ........................................................................................ 94
Figure 4.3. Four 2D IP reconstructed permittivity (top) and conductivity (bottom) images
(1300 MHz) of the phantom shown in figure 39, the phantom’s properties at 1300 MHz
are given in table 4.1. ........................................................................................................ 95
Figure 4.4. Experimental phantom set up showing (a) breast phantom with irregular
shaped inclusion placed in imaging tank (b) top-down view of phantom showing breast
and tumor phantom materials............................................................................................ 98
Figure 4.5. Reconstructed images (1500 MHz) of the phantom shown in figure 41 using
varying amounts of long-range data: (a) permittivity 2 XP, (b) conductivity 2 XP, (c)
permittivity 4 XP, (d) conductivity 4 XP, (e) permittivity 9 XP and (f) conductivity 9 XP
........................................................................................................................................... 99
Figure 4.6. TX-RX plane separation distances for (a) TX and farthest in-plane RX
antenna element, (b) TX and farthest out-of plane RX antenna element, (c) TX and
closest in-plane RX antenna element and (d) TX and farthest out-of plane RX antenna
element out-of-plane. ...................................................................................................... 102
Figure 4.7. HFSS (Canonburg PA, USA) simulated electric field overlays of the TX
antenna elements for (a) TX and farthest RX in-plane element, (b) TX and farthest RX
out-of-plane antenna element, (c) TX and closest RX in-plane antenna element and (d)
TX and closest RX out-of-plane antenna element. ......................................................... 103
Figure 4.8. Simulated S21 measurements for (a) TX and farthest RX antenna element for
various plane-separation distances (b) TX and closest RX antenna element for various
plane-separation distances. ............................................................................................. 105
xv
Figure 4.9. Schematic of experimental set-up showing the transmitting element (green),
ineligible IP elements (red), utilized receiving channels (blue) and their symmetric
counter parts (white). ...................................................................................................... 109
Figure 4.10. Reconstructed permittivity images of the phantom shown in figure 41 (a) at
1500 MHz using FD and (b) at 1300 MHz using FD. .................................................... 116
Figure 5.1. New motion system highlighting (a) placement of rotary motor-controlled
linear-actuators and motion encoder placed in data acquisition structure and (b) flash
programmable amplifier drives and AC/DC power converter. ....................................... 118
Figure 5.2. CME2 software GUI utilized for developing motion sequences that are stored
in the amplifier’s flash memory drive. This image shows more than 24 different motion
sequences, each of which can be executed through the CME2 software or an ASCII
command exporter. ......................................................................................................... 123
Figure 5.3. Kvasier Leaf Lite HS Controller-Area Network (CAN) to Universal Serial
Bus (USB) adapter (Mission Viego, CA). ...................................................................... 126
Figure 5.4. New motion system’s LABVIEW user interface showing a 7x7 motion
combination matrix. ........................................................................................................ 128
Figure 6.1. MR compatible MT imaging chamber’s (a) antenna array connected to semiridged coaxial cable and (b) placed on the MRI system’s patient positioning table. ..... 131
Figure 6.2. Photographs of (a) MT system’s electronics attached to wall mounted
bulkhead panel separating shielded MRI room from control room, (b) MT system’s
mounted SMA adapters, (c) wall mounted bulkhead panel inside MRI suit and (d) MR
compatible MT imaging array positioned inside the bore of the MRI system. .............. 132
xvi
Figure 6.3. Photograph of newly developed MT imaging prototype showing mounted
optical scanner. ............................................................................................................... 134
Figure 6.4. (a) Hardware for the updated motion control system and optical scanning
system placed in electronics cart underneath the new patient bed and (b) new illumination
apparatus and associated hardware placed next to current microwave electronics’ system.
......................................................................................................................................... 136
Figure 6.5. Master computer system’s GUI ready for the user to enter examination
parameters. ...................................................................................................................... 137
Figure 6.6. Slave computer system’s GUI prior to the establishment of communication
with the Master system. .................................................................................................. 138
Figure 6.7. Slave computer system’s GUI after the establishment of communication with
the Master system. .......................................................................................................... 140
Figure 7.1. Simulation meshes used for calculating the forward data using a FEM method
described in [##]: (a) single, (b) two and (c) three inclusion cases showing the
background (dark blue), breast (light blue) and inclusion(s) (green, orange and red)
regions, respectively. ...................................................................................................... 145
Figure 7.2. Reconstructed (a) conformal and (b) uniform permittivity images of the
simulation set-up shown in figure 7.1a. .......................................................................... 146
Figure 7.3. Reconstructed (a) conformal and (b) uniform conductivity images of the
simulation set-up shown in figure 7.1a. .......................................................................... 147
Figure 7.4. Reconstructed (a) conformal and (b) uniform permittivity images of the
simulation set-up shown in figure 7.1b. .......................................................................... 149
xvii
Figure 7.5. Reconstructed (a) conformal and (b) uniform conductivity images of the
simulation set-up shown in figure 7.1b. .......................................................................... 149
Figure 7.6. Reconstructed (a) conformal and (b) uniform permittivity images of the
simulation set-up shown in figure 7.1c. .......................................................................... 152
Figure 7.7. Reconstructed (a) conformal and (b) uniform conductivity images of the
simulation set-up shown in figure 7.1c. .......................................................................... 152
Figure 7.8. (a) Schematic of the experimental imaging domain and (b) a photograph of
the phantom placed in the imaging system. .................................................................... 154
Figure 7.9. Reconstructed Permittivity Images using a (a) 7 cm and (b) 5.5 cm radius
reconstruction mesh ........................................................................................................ 155
Figure 7.10. Reconstructed permittivity images of the phantom shown in figure 7.7 for a
variety of parameter meshes: (a) 7 cm, (b) 6 cm, (c) 5.5 cm (exact) (d) 5 cm and (e) 4.5
cm radii, respectively. ..................................................................................................... 158
Figure 7.11. Plot of the inclusion’s permittivity as a function of OUT-parameter mesh
mismatch. ........................................................................................................................ 159
Figure 7.12. Reconstructed permittivity images of the phantom shown in figure 7.7 for a
variety of horizontal translations (top) +5 to +25 mm translations in 5 mm increments,
middle exact conformal and (bottom) -5 to -25mm translations in 5 mm increments. .. 160
Figure 7.13. Schematic of the imaging domain used for the 3D simulations shown in
figure 7.14 including a bath (light blue), breast (dark blue) and tumor (red) regions. ... 161
Figure 7.14. 1300 MHz reconstructed conductivity images (from simulated data
corresponding to the imaging domain shown in figure 7.12) using varying mesh heights
xviii
and radii: (a) 7 cm radius with a 4 cm height, (b) 5.75 cm radius with a 3.5 cm height, (c)
5.0 cm radius with a 3.25 cm height (exact) and (d) 4 cm radius with a 3 cm height. ... 162
Figure 7.15. Photographs of (a) plastic breast mold (b) submerged in the imaging
system’s illumination chamber. ...................................................................................... 163
Figure 7.16. Scanner extracted (a) point cloud and (b) wire mesh of the breast phantom
shown in figure 7.15. ...................................................................................................... 164
Figure 7.17. CMI reconstructions (1300 MHz) of the breast phantom shown in figure
7.15 (a) permittivity and (b) conductivity. ...................................................................... 165
Figure 7.18. Uniform mesh reconstruction utilizing the same data as the reconstructed
images in figure 7.17....................................................................................................... 165
xix
List of Tables
Table 3.1. Tabulated SNR [dB] for each channel of the system’s parallel detection
scheme............................................................................................................................... 78
Table 4.1. Measured dielectric properties of the phantom shown in figure 39. ............... 94
Table 4.2. Measured dielectric properties of the phantom shown in figure 4.4. .............. 98
Table 4.3. Transceiver unit’s receiving channel’s electronic parameter values ............. 104
Table 4.4. Tabulated SA sensitivity results for the far neighbor (Channel 10) .............. 112
Table 4.5. Tabulated SA sensitivity results for the middle neighbor (Channel 12)........ 112
Table 4.6. Tabulated SA sensitivity results for the near neighbor (Channel 14) ............ 113
Table 7.1. Position and property values of the regions for the three simulation cases
(shown in figure 7.1). ...................................................................................................... 145
Table 7.2. Recovered permittivity and conductivity values for the one inclusion case,
including error calculations for the uniform (top) and conformal (bottom) images,
respectively. .................................................................................................................... 146
Table 7.3. Recovered permittivity and conductivity values for the 1st inclusion of the twoinclusion case (green-inclusion in figure 7.1b). .............................................................. 148
Table 7.4. Recovered permittivity and conductivity values for the 2nd inclusion of the
two-inclusion case (orange-inclusion in figure 7.1b). .................................................... 148
Table 7.5. Recovered permittivity and conductivity values for the 1st inclusion of the
three-inclusion case (green-inclusion in figure 7.1c). ..................................................... 151
Table 7.6. Recovered permittivity and conductivity values for the 2nd inclusion of the
three-inclusion case (orange-inclusion in figure 7.1c). ................................................... 151
xx
Table 7.7. Recovered permittivity and conductivity values for the 3rd inclusion of the
three-inclusion case (red-inclusion in figure 7.1c). ........................................................ 152
Table 7.8. The DPs and positions of the phantom materials shown in figure 7.7b. ....... 155
Table 7.9. Recovered permittivity values and the corresponding % error over the breast
region for the phantom shown in figure 7.8. ................................................................... 156
Table 7.10. Position, permittivity values and error calculations of the phantom shown in
figure 7.7 for a variety of parameter meshes (7 cm represents the uniform mesh and 5.5
cm represent the exact boundary conformed mesh, respectively). ................................. 157
Table 7.11. Position and conductivity properties of the simulated imaging domain shown
in figure 7.13. .................................................................................................................. 161
Table 7.12. Dielectric properties of the breast phantom and tumor inclusion at 1300
MHz…………………………………………………………………………………….164
xxi
List of Acronyms
Actuator-Motor
AM
Breast Cancer
BC
Breast Imaging Reporting and Data Systems
BI-RADS
Computed Tomography
CT
Conformal Microwave Imaging
CMI
Controller-Area Network
CAN
Cross-Plane
XP
Data Acquisition System
DAQ
Dielectric Property
DP
Dielectric Property Assessment
DPA
Direct Current
DC
Electromagnetic
EM
Electromagnetic Energy
EME
Electronic Impedance Tomography
EIT
Finite Difference Time Domain
FDTD
Full Data
FD
Gain
G
Gauss Newton
GN
Graphical User Interface
GUI
In-Plane
IP
Input Noise
IN
Input/Output
I/O
xxii
Intermediate Frequency
IF
Local Oscillator
LO
Magnetic Resonance
MR
Magnetic Resonance Imaging
MRI
Microwave Imaging
MI
Microwave Imaging Spectroscopy
MIS
Microwave Tomographic Imaging
MTI
National Institute of Health
NIH
National Instruments
NI
Near Infrared
NIR
Noise Figure
NF
Noise Floor
NF
Object Under Test
OUT
Out-of-Plane
OOP
Personal Computer
PC
Power Divider
PD
Radio Frequency
RF
Radio Frequency Signal Generator
rfSG
Receiver
RX
Region of Interest
ROI
Signal to Noise Ratio
SNR
Single-Pole-Double-Throw
SPDT
Single-Pole-Four-Throw
SP4T
xxiii
Single-Pole-Single-Throw
SPST
Sub-Array
SA
Three-dimensional
3D
Time-Harmonic Electromagnetic
THEM
Tissue-Sensitive-Adaptive-Radar
TSAR
Transistor-Transistor Logic
TTL
Transmitter
TX
Two-dimensional
2D
Ultra-Wideband
UWB
Vector Network Analyzer
VNA
Virtual Instrument
VI
Voltage Standing Wave Ratio
VSWR
xxiv
1. Introduction
The interaction of electromagnetic energy (EME) with biological tissue has been
thoroughly investigated for several decades [1, 2, 3, 4, 5]. Results from these
investigations highlight the value of using EME detection systems to assess tissue health.
It has been shown that microwave radiation can identify dielectrically contrasting regions
within biological materials [6, 7, 8, 9]; an endogenous contrast exists between the
electrical properties (permittivity and conductivity) of normal and abnormal breast tissue,
and as a result microwave radiation can be used to identify and assess abnormal regions
within the breast volume. The recovery of accurate property measurements in clinically
feasible time periods is requisite to the success of any imaging modality. Consequently,
this thesis will focus on the development and advancement of a motion optimized
tomographic microwave imaging (MI) system for applications related to breast cancer
(BC) detection and diagnosis.
MI systems are characterized as passive, hybrid or active systems. The differences
between these classification schemes are related to the generation of the measured signal
and its detection strategy. Independent of the detection technique or the physiological
phenomena responsible for signal emission, all of these microwave detection techniques
exploit the observable property contrast that exists between normal and abnormal breast
tissue for clinical applications.
Passive approaches seek to monitor the naturally emitted microwave signals that
are generated through thermal-vibrations of tissue components [10]. This technique
potentially plays a role in breast cancer detection; the higher temperature of the abnormal
tissue relative to the normal surrounding tissue results in an increased amount of thermal
1
vibration and an increased amount of associated microwave energy emission [9, 10]. This
allows the detection of pathological conditions where disease-related temperature
differentials exist; such is the case for malignant and normal breast tissue [12, 13, 14, 15,
16, 17, 18]. The regional contrast in signal emission can be passively detected using
radiometric measurement techniques [19, 20, 21, 22, 23, 24]. However, depth-dependent
signal attenuation makes the detection of abnormalities located deep within the breast
volume difficult. This phenomenon has resulted in the development of hybrid and active
systems capable of strengthening detected property-dependent measurement signals.
In hybrid systems, thermal expansion is actively induced using microwave
signals with the associated pressure waves detected using ultrasound [25, 26, 27, 28].
The increased conductivity of a tumor region relative to the surrounding healthy tissue
creates an environment that is more susceptible to thermal expansion effects. The contrast
between the emitted pressure waves of the normal and abnormal regions can be detected
using an acoustic measurement system [29]. The combined system exploits ultrasound’s
high resolution imaging capabilities while enhancing its relatively low image-contrast
using the increased strength of the thermally induced emission signal.
Active approaches [30, 31, 32, 33, 34, 35] utilize a transmitted microwave signal
for tissue interrogation, and some form of the reflected or transmitted signal for dielectric
property assessment (DPA). Developing an active MI system requires a combination of
numerical modeling and hardware development. Due to this relationship, several active
MI strategies have been investigated, all varying in their numerical modeling and
hardware sophistication. These methods include ultra-wideband radar-based confocal
techniques and tomographic imaging approaches.
2
Confocal approaches seek to localize high scatter regions within the breast
volume using monostatic [36] or multistatic [37] radar-based techniques. A combination
of forward and backscatter signal measurements are used to generate microwave energy
scatter maps of the breast volume. High scatter areas correspond to dielectrically
contrasting regions, which may be indicative of a breast abnormality. These techniques
have progressed beyond simulation investigation; credible systems capable of
experimental phantom-based imaging have been developed, with the more advanced
systems nearing the clinically investigation stage [36, 37]. This approach utilizes scatter
parameter (S-parameter) measurements obtained with the use of a vector network
analyzer (VNA); the acquired frequency-domain data is transformed to the time-domain
and used for tissue analysis. Issues related to ultra-wideband (UWB) techniques do exist,
and will be discussed in section 1.4.1.
Alternatively, microwave tomographic (MT) approaches transmit microwave
energy through the breast volume and utilize signals detected around the volume’s
perimeter to recover its dielectric properties (DPs). This requires the use of iterative
numerical inversion algorithms that attempt to minimize the difference between the
measured and calculated fields. Methods utilizing single-frequency [38], multi-frequency
[39] and time-domain [40] data have been explored.
Initial investigations were predominately simulation-based; however, efforts have
resulted in the development of a variety of actual tomographic prototypes [40, 41, 42,
43]. Although the majority of these systems are currently at the phantom verification
stage, more advanced systems have successfully translated to the clinic for applications
related to BC [41]. While the inversion technique is computationally expensive,
3
especially for volumetric three-dimensional (3D) image reconstruction, advances in
computational power and the use of parallel computing techniques have made it practical.
A more recent advance by Grzegorczyk et al [44] using a discrete dipole approximation
has reduced the computational time for 3D reconstructions to fewer than ten minutes. A
variety of time-domain MT systems have been investigated, the majority of which utilize
VNAs for signal emission and detection; as a result these systems are associated with low
signal to noise ratio (SNR). Frequency-domain MT systems can overcome this issue by
using a signal generation source with a larger dynamic range; consequently, our
frequency-domain MT system has employed this strategy to specifically overcome this
issue.
As is the case with the development of any technology, the construction of a
clinical MI system involves design trade-offs related to hardware selection and cost,
signal attenuation and detection strength, operating frequency and resolution, and antenna
selection and array movement [45]. Accordingly, successful system materialization
requires the combination and implementation of hardware and software strategies that
can effectively operate over the desired frequency range while producing reliable
property information that is clinically relevant.
UWB technologies’ low SNR and the signal processing techniques needed to
eliminate undesired signal scatter, combined with advances in computational power and
available hardware, have influenced our decision to advance a multi-frequency, active
tomographic MI system. Henceforth, the work in this thesis will focus on the
development of our current (3rd generation) clinical MI system, including a thorough
4
analysis of its operational performance and significance as a clinical tool for applications
related to BC.
5
1.1 Motivation for developing alternative breast cancer imaging modalities
A major emphasis has been placed on the advancement of BC detection and
diagnostic systems in recent years [46]. Current research indicates that in the United
States, 1 in 8 women are expected to develop a type of invasive BC over their lifetime
[47]. Estimates suggest 39,520 BC related deaths in 2011, with approximately 226,870
new cases of invasive BC expected for 2012 [47, 48]. According to American Cancer
Society statistics, maximum survival rates are strongly associated with early detection;
the 5-year survival rate of a BC patient decreases from 93% to 15% depending on the
stage at which the cancer was detected [49]. The risk of developing BC is a function of
genetic, environmental and tissue compositional factors; as a result of these issues, it is
recommended that women with a 20% or higher lifetime risk of developing BC undergo
routine breast screening every year [49].
Unfortunately, currently accepted screening techniques such as X-ray
mammography use ionizing radiation, require breast compression, and suffer from
sensitivity and specificity limitations due to low soft-tissue contrast. Radiation induced
breast mutations account for 9-32 BC cases per 10,000 women screened [50].
Overdiagnosis resulting in the treatment of insignificant abnormalities accounted for 33%
of mammographically detected BCs in 2011 [51], with recent estimates indicating the
overdiagnosis of more than 1,000,000 women since the introduction of routine
mammographic screening [52]. Moreover, it has been shown that false-positive and
false-negative rates for X-ray mammography can reach as high as 50% [53] depending on
breast tissue density [54, 55] and cancer growth rates [56].
6
These shortcomings indicate the need for alternative means of accurately
assessing breast tissue health, and have prompted the development of novel alternative
BC imaging techniques. In fact, the National Institute of Health (NIH) has recently
placed more emphasis on developing technologies that will reduce the amount of ionizing
radiation used in clinical imaging modalities such as X-ray, computed tomography (CT),
fluoroscopy and nuclear medicine [57]. These new systems are innovative in their ability
to extract property information from previously underutilized EM signals. Driven by the
desire to develop EME detection systems that can overcome the sensitivity and
specificity limitations of current clinical imaging systems, a variety of self-contained and
multimodal alternative BC imaging systems have been developed; capable of
independently assessing tissue health with the potential to compliment additional imaging
technologies.
Dartmouth College has spearheaded the development of EM model-based
alternative imaging techniques for clinical applications related to BC. Three of these
techniques: electronic impedance tomography (EIT), near infrared (NIR) imaging and MI
spectroscopy (MIS) are currently utilized in clinical trials at Dartmouth Hitchcock
Medical Center (Lebanon, NH). Each system operates over a specific non-ionizing
frequency range, and is capable of extracting relevant diagnostic information using
custom designed hardware and software.
Unfortunately, for non-invasive breast tissue assessment using these alternativeimaging techniques, EIT and NIR are associated with limited resolution and increased
depth-dependent signal attenuation, respectively. Additionally, the non-contacting nature
of the MT system increases patient comfort when compared to the physical contacting of
7
the electrodes and optical fibers required for EIT and NIR imaging. These issues have
motivated our development of an advanced MT system.
Studies have indicated that in the microwave frequency range, the contrast
between the DPs of healthy and malignant breast tissue is larger than other anatomical
features [9, 58, 59]. MIS has the ability to detect a wide range of DP values and due to
the contrast that exists between the DPs of normal and abnormal breast tissue, MIS has
shown strong promise as an alternative BC imaging modality. The limitations of current
traditional imaging techniques, in conjunction with MIS’ ability to non-invasively detect
a wide range of DP values have been influential factors in investigating MT’s relevance
as a biomedical tool.
8
1.2 Physical basis underlying the use of electromagnetic energy for dielectric
property assessment
The interaction of EME with biological tissue is governed by a combination of
physiological mechanisms [5, 59, 60, 61]. Applying a time-harmonic EM (THEM) field
to these non-magnetic materials generates electric and magnetic forces that act on
molecular and cellular components possessing a net electric charge and/or a dipole
moment. Electrical properties are a measure of a material’s ability to interact with an
applied THEM field; this interaction is frequency dependent and is influenced by various
factors such as tissue composition and vascularization [5]. DPA using EM excitation is a
function of a material’s permittivity and conductivity characteristics.
Molecules possessing an electric dipole moment will experience charge
displacement due to time-varying field generated EM forces; this displacement
mechanism, known as polarization, opposes the force of molecular attraction [62].
Permittivity is a measure of a materials ability to become polarized in the presence of an
applied field, and is indicative of energy storage [62]. Additionally, this EM interaction is
associated with the movement o0.
f free charges and dielectric heating within the tissue. Conductivity is a measure
of the energy dissipation and charge movement characteristics within a material under the
influence of an applied THEM field [62]. These material dependent interactions and the
corresponding material dependent electrical property measurements can be used as
clinical tools for detecting and diagnosing various physiological conditions [5].
9
A THEM field will excite a steady-state system, generating charge movement and
inducing a current density ( )
within a biological material. Using Maxwell’s
Laws, the frequency dependent dielectric response can be expressed as:
(
where
is the electric field
,
)
is the conductivity of the material
angular frequency of the applied field and
is √
,
is the complex permittivity, with
representing its real and imaginary parts, respectively.
and
(1.1)
is the
and
is the permittivity of free space
[63]. In the time domain, the reestablishment of system
equilibrium after EME induced excitation is associated with a material dependent
relaxation time ( ). Biological materials are characterized using multiple relaxation
processes and times due to their heterogeneous nature [63]. In conjunction with the
induced , the application of a THEM field at time
charge densities (
)
will induce material dependent
within the biological material, each with a
corresponding . This relationship is expressed as:
(
where
⁄
)
⁄
(
)
(1.2)
represents the instantaneous surface charge density response due to electronic
polarization. System response in the frequency domain can be obtained by applying a
Laplace transform to equation 1.2 [64], and is expressed as:
10
(
where
(1.3)
)
represents an experimentally based parameter, and
and
represent the high
and low frequency permittivity manifestations, respectively [65].
Using Kramers-Kronig relations for materials with multiple relaxation times,
and
can be represented as [65]:
( )
( )
where
( )
∫
(1.4)
( )
∫
(1.5)
is a real variable of integration. As noted in [65], these relationships are useful
for interpreting the frequency-dependent behavior of dielectrics based on dielectric
property measurements. From these equations it can be seen that the frequency response
of
entirely determines that of
parts of equation 1.3,
and
can be expressed as:
(
(
and vice versa [61]. By equating the real and imaginary
)
(
(1.6)
)
)
(
)
(
(1.7)
)
11
where
and
are known as the dielectric constant and dielectric loss factor,
respectively.
For EM waves traveling in a lossy medium,
and
can be stated in terms of
the material’s electrical properties, such that:
(1.8)
(1.9)
where
and
are the relative permittivity and conductivity of the interrogated material,
respectively.
As identified in the Kramers-Kronig relation equations, the existence of
implies a frequency dependent
. When these properties are dispersive, the frequency
Figure 1.1. Dispersion relations of
(compliments of science.direct.com).
and
12
for typical a biological tissue
dependence of the measured permittivity and conductivity are governed by four
physiological mechanisms known as
,
and
dispersions [5, 60, 61, 66]. These
dispersion relations link the DPs such that a decrease in permittivity is associated with an
increase in conductivity, and vice versa [59]. The dispersive nature of these properties,
along with their underlying frequency-dependent mechanisms, is illustrated in figure 1.1.
dispersions occur over the Hz – kHz range. Mechanisms responsible for this
dispersion include counter ion effects, ionic diffusion, membrane effects and the charging
of intercellular structures. For a suspension of cells with 30
radii and a volume
fraction of 0.9, the theories of Schwartz [67, 68] and Grosse [69, 70] predict a dielectric
increment of roughly
units. This result is consistent with the measured permittivity of
tissues at audio frequencies, which is typically in the range from
[59].
Assuming a relaxation frequency of 100 Hz, calculations using the Kramers-Kronig
relations yield a total conductivity increase of ~0.005 S/ , indicating that tissues are
resistive at low frequencies, despite their large permittivity values [5].
dispersions occur over the high kHz – mid MHz range [5, 60, 61, 71, 72, 73].
This dispersion mechanism includes the capacitive charging of cellular membranes and
the dipolar reorientation of tissue proteins. Blood, with a dielectric increment of 2000
units and a relaxation frequency of 3 MHz [5] yields a total conductivity increase of ~0.4
S/ , as specified by the Kramers-Kronig relations. For tissues, the static permittivity and
relaxation times of this dispersion mechanism are typically larger than that of blood [5].
dispersions [5, 60, 61, 71, 72, 73], originally termed by Grant [5], were first
noted in hemoglobin [74, 75] and later examined in albumin [76]. Occurring over the 0.1
– 3 GHz range, mechanisms responsible for this dispersion include the dipolar relaxation
13
of bound and free water, the rotational relaxation of polar side chains and the diffusion of
counterions along small regions of charged surfaces. In this frequency range soft tissues
typically experience a relative permittivity decrease of 10 – 20 units, resulting in a
conductivity increase of ~0.5 S/
[5].
dispersions occur over the 0.1 – 100 GHz range [5, 60, 61, 71, 72, 73]. This
dispersion mechanism includes the dipolar relaxation of polar media and proteins. Due to
tissue’s high water content, at body temperature, this dispersion occurs with a center
frequency near 25 GHz. Hydrated soft tissues typically have a dielectric increment of 50
units; using a relaxation frequency of 25 GHz yields a total conductivity increase of ~70
S/
[5].
In the microwave frequency range, the DPs of biological tissues are dominated by
dipolar relaxation of water, which accounts for roughly 80% of the volume of most soft
tissues [5]. Using previously published data on tissue DP values [77] and water content
[78], Dr. Tian Zhou investigated the relationship between tissue DPs and water content at
1, 2 and 5 GHz. His analysis showed a nearly linear relationship between the properties,
with a slight nonlinear deviation in the conductivity pattern of low water content tissues
at lower frequencies. This aberration is driven primarily by the presence of bound water,
which decreases with increasing frequency [79].
These findings are consistent with the data reported in [80]. Results from this
analysis reveal two key water-dependent tissue dielectric property features:
1. Tissue permittivity appears to be proportional to the overall tissue
water content
14
2.
Tissue conductivity is directly related to the proportion of free water
at frequencies below 2 GHz.
It has been shown that tumor tissue has greater water content than normal tissue of the
same type [82, 83]. Based on the relationship between tissue DP values and tissue water
content, tumor tissue is expected to have higher DP values when compared to its
corresponding normal counterpart. This is supported by the findings in [84], where the
contrast in the DPs of tumor and normal adipose-dominated tissues in the breast at
microwave frequencies can be as large as 10:1.
15
1.3 Breast tissue dielectric properties
Normal breast tissue is heterogeneous by nature, and as a result literature
regarding breast tissue categorization is rather ambiguous [85]. Consequently this
analysis will focus on three primary breast tissue categories: adipose, glandular and
connective tissues, including benign and malignant breast abnormalities. Breast tissue
density is strongly associated with BC risk; variations in breast tissue composition affect
its density. The Breast Imaging Reporting and Data System (BI-RADS) define four
radiographic breast densities: predominately fat, scattered dense, heterogeneously dense
and extremely dense [85]. This compositional variance, combined with tissue temperature
variation and water content changes due to physiological conditions such as
menstruation, pregnancy and lactation result in a large range of DPs for both healthy and
abnormal breast tissue [86].
Despite these DP variations, their use in assessing breast tissue health is still
clinically relevant. A collection of normal and malignant breast tissue dielectric property
Figure 1.2. Summary of measured DPs for normal (black) and malignant (red) breast
tissue at radio and microwave frequencies.
16
measurements are summarized in [87]. Figure 1.2 shows the corresponding dielectric
property data for normal (black) and malignant (red) breast tissue published by several
research groups. An identifiable contrast in the DPs of healthy and malignant breast
tissue can be seen over the indicated frequency range, notwithstanding the tissue’s
compositional heterogeneity and associated DP variations. Our decision to advance a
system that operates from 500 MHz – 3 GHz is a result of the large contrast in the
permittivity and conductivity values over that range.
Poplack et al [88] quantitatively assessed the DP contrast in wormen with
abnormal findings that were identified using conventional breast imaging techniques. A
total of 80 subjects in the abnormal group and 50 normal control subjects were examined
using Dartmouth’s MT system. Of the 80 abnormal subjects, findings were clasified as
malignancies (26), fibrocystic (41), fibroadenoma (eight) and other benign abnormalities
Figure 1.3. Bar charts of the mean imaged DP property contrast ratios using
Dartmouth’s MI system. Vertical lines represent the 95% confidence interval, and
mean numerical value inside each box. P values appear above the bars; cancers have
the highest mean contrast ration relative to benign breast abnormalities and normal
control subjects.
17
(five). Mean permittivity and conductivity ratios of the region of interest (ROI) to
background are summarized by diagnostic group (figure 1.3). For lesions with a
mamographic identified abnormality that was greater than 1 cm, differences in the mean
conductivity contrast were highly significant (P < 0.002) in group comparisons for the
diagnostic categories.
These findings, along with additional data published in [85] show consistency
regarding the diagnostic value of breast tissue DP data, which can be summarized as:
1. The mean conductivity of normal breast tissue is less than 1.5
for frequencies up to 3.5 GHz
2. Malignant breast tissues have higher mean permittivity and
conductivity values than normal breast tissue
3. Adipose tissues have the lowest mean permittivity and
conductivity values
4. The low conductivity values of normal breast tissue enables
penetration of microwave frequencies up to the GHz range
5. At frequency ranges of 1 GHz – 3 GHz, DPs can help classify
normal and malignant breast tissue
Although variations in the DPs of both normal and abnormal breast tissues exist, results
from [9, 82, 85, 87] support the idea that microwave technologies are useful tools for
assessing tissue health; their role is clinically relevant, especially for BC diagnosis [88]
and therapy monitoring applications [89].
18
1.4 Dielectric contrast imaging using microwave energy
Systems capable of measuring
and
allow the calculation of a material’s
permittivity and conductivity using equations 1.8 and 1.9. Joines et al [58] measured the
electrical properties of various tissue types with a flat-ended coaxial cable and VNA
using an admittance model given by:
(
)(
)
where Y is the measured admittance,
(1.10)
is the angular frequency,
measured complex permittivity with relative permittivity
√
. The (
is the
and conductivity , and j =
) factor is the effective area-to-distance ratio, and can be calculated
theoretically [90] or calibrated by measuring a material with known properties [91].
Figure 1.4. Power absorption ratio plots of malignant to normal tissue for various
tissue types.
19
Figure 1.4 plots the power absorption ratio of malignant (Pm) to normal (Pn) tissue for
various tissue types, and is defined as:
(
(
⁄
⁄
)
)
(
(
)
⁄
⁄
) (
⁄
)
(1.11)
where the subscripts m and n correspond to the malignant and normal tissue’s electrical
property values, respectively. Over the indicated frequency range, the largest contrast in
the measured electrical properties occurs in mammary tissue. Although taken ex vivo,
these measurement results are indicative of the overall trend associated with using
microwaves to assess breast tissue health; a discernible contrast exists between the
electrical properties of malignant breast tissue when compared to its healthy counterpart,
and this contrast can be exploited for clinical applications related to BC.
Admittance and probe based S-parameter measurement techniques are useful for
assessing the DPs of breast tissue at specific points on a tissue’s surface; however,
volumetric analysis using this approach is rather difficult. Measuring the DPs of a large
region of interest requires multiple measurements taken at numerous locations. This
necessitates probe or material movement during the measurement process, which can
introduce undesired measurement errors related to non-optimal tissue-probe contact and
cable tension/flexion. Additionally, these techniques have limited depth measurement
capabilities when noninvasive property assessment is desired. The ability to assess breast
20
tissue health non-invasively has fostered the development of several detection strategies,
and has further influenced our decision to develop an advanced non-invasive MT system.
21
1.4.1. Ultra-wideband and space-time imaging techniques
Ultra-wideband (UWB) radar-type approaches seek to localize regions of
significant backscatter energy resulting from dielectrically contrasting regions within the
breast, e.g., normal and malignant breast tissue. Reflected (S11) and transmitted (S21)
signal measurements can be used to generate microwave backscatter intensity maps of the
irradiated region. In these images, high intensity zones correspond to tissue areas where
increased scattering has occurred, indicating regions
with
higher
and/or
Clinical
Setup
and permittivity
Results
conductivity, which may
presence
of an abnormality.
The suggest
clinical the
setup
and measurement
procedure was designed to replicate the phant
System overview
setup as closely as possible, as follows. Firstly, the curved array is filled with
Klemm et al [37] have developed a mulitstatic UWB radar-based microwave
matching
medium.
ThenThis
a spherical
(electrical parameters the same as for
The developed system is essentially the UWB
microwave
radar.
radar usesshell
a real
aperture array of UWB antennas and operates
in medium,
a multi-static
mode.
Antennastransparent)
are
matching
and hence
essentially
is placed 2cm from the antennas
imaging
where conforming
the scatteredwell
waveforms
are collected
and time-shifted to focus at
positioned on a section
of thesystem
hemi-sphere,
to
the
curved
breast
shape.
small amount of the matching medium is poured into the shell, to provide good cont
For the detailed description of the hardware as well as post-reception focusing algorithms
between
patient’s
breastThis
and aresults
shell and to avoid any air gaps. Then, the patient l
individual pixels
within
the[6]afield
of view.
please refer to [5] (first-generation
prototype)
and
(second-generation
prototype).in the generation of useful
down on a couch and the array is mechanically positioned with the breast comforta
on
the sphericalscatter.
shell. A
photo
of the clinical
setup with
Experimental
Setup and
Results
microwave
data for positioned
regions
with
significant
Their
radar-based
system’s
breasta patient (actuall
healthy volunteer) in the correct position is shown in Figure 3.
During laboratory experiments, the array is first filled with the matching medium, the
interface and symmetrically curved 16 element antenna array (figure 1.5a) is shown with
spherical skin phantom is placed in the correct position, and then we attach a tank to the
top of the antenna array to finally fill it with the breast fat equivalent liquid (the same as
the matching medium). This setup represents truly three-dimensional (3D) breast
phantom. The chest wall is not considered in our experiments.
Figure 1. Array, feed and switching (phantom
would
normally beradar-based
on top)
(a)
(b) for breast cancer detection. Patient ly
Figure
3. Microwave
clinical setup
Figure 1.5. (a) University
Bristol’s UWB imaging system highlighting their
in a proneofposition.
In Figure 2 we present an example of the detected 8mm spherical phantom tumor. The
backscatter interface and symmetrically curved antenna array and (b) patient lying
obtained image is clear with little clutter and this is typical of our phantom images. More
prone
breast inserted
in imaging
system.
experimental results using
thewith
developed
system
be
foundsetup
in [5,described
6].
Using can
the
clinical
above, we have performed a small scale blind clini
trial with real breast cancer patients at the Bristol Oncology Center. In Figure 4 below
present the mammogram of the post-menopausal woman with a breast cancer, and
corresponding image using the developed radar-based microwave system. T
microwave system has detected the tumor and in the correct position. The estimate of
22
tumor position between both imaging techniques cannot be precise however, as the bre
during the mammography measurement is strongly compressed.
a patient lying prone, with her breast inserted into the system (figure 1.5b). The visible
tumor mass highlighted in the X-ray mammogram (figure 1.6a) was detected by the
UWB system, and is represented as the high intensity region in the corresponding
backscatter intensity map (figure 1.6b). This system is capable of detecting tumor-like
inclusions less than 1 cm in diameter.
(a)
(b)
Figure 1.6. (a) X-ray mammogram highlighting abnormal region and (b) backscatter
intensity map showing abnormal region in red.
Recent efforts to increase their system’s imaging performance have resulted in the
development of a new experimental system with enhanced imaging capabilities.
Significant improvement has been achieved through the creation of an updated patient
interface that incorporates 31 newly designed wide-slot UWB antennas [38]. This
experimental system is capable of detecting 7 mm diameter tumor-like inclusions in
homogeneous breast phantoms and low contrast conditions. The system collects
frequency-domain S21 parameter measurements with a VNA (VNA; Rhode & Schwarz
ZVB20), and uses a post-processing step to transforms the 456 independent
23
measurements into the time-domain. The computer controlled VNA and switching-bank,
along with the new patient interface is shown in figure 1.7.
KLEMM et al.: MICROWAVE RADAR-BASED DIFFERENTIAL BREAST CANCER IMAGING
a background measurement, it could
narios with breast cancer patients [22]
A modified delay and sum (DAS)
used to form 3D images of scattered
scattered energy at the given focal
volume, can be expressed as
Figure 1.7. University
of Bristol’s
new system.
experimental UWB imaging
system
Fig. 3. Experimental
setup of our imaging
where
( is the n
highlighting the VNA, switching-bank and modified patient interface. array),
is the location dependent w
pre-processing,
is the measured ra
breast tissue equivalent liquid [33] has a relative dielectric con- time-delay. is the length of the integr
stant of about 10 and an attenuation of 1.2 dB/cm at 6 GHz. This
and
were introduced
material
is
also
dispersive,
and
its
frequency-dependent
charactennas’
effects
and
they
UWB radar-based techniques only attempt to localize and characterize high are defined as
teristics have been presented in [34].
As an alternative to the lossy matching liquid, we have rescatter regions within
volume;
they
not between
attemptantennas
to recover the associated
cently the
builtbreast
a ceramic
shell to fill
the do
distance
and a breast skin (a 1 mm thin layer of matching liquid still needs
DPs. The literature
suggests
that antennas
UWB techniques
canshell).
localize
abnormalities
to be
used between
and the ceramic
Also, to
be
where: [87]isand
a vector beginning
able to accommodate breasts of different size, ceramic insert ending at the center of curvature [
shells have
been made.
The ceramic matching
shell and lesion
inserts discrimination
differentiate malignant
and benign
abnormalities
[87]; however,
and
Fig. 2(a)],
is
a vector beginning
were custom designed, using low loss material with controlled ending at a focal point
dielectric constant
material Eccostock HiK500F from is the maximum assumed antenn
characterization utilizing
S-parameter measurement
data necessitates the exploitation of
Emerson & Cuming. All ceramic parts are shown in Fig. 2(b).
tially, (2) checks whether for the
Next, a curved skin phantom was developed. The skin layer is point
morphology-dependent
signal
characteristics,
which
may[shown
be in
difficult
when the
2 mm thick,
it is a part
of a 67 mm-radius
hemisphere
sumed antenna beamwidth (Tx- or R
Fig. 2(a)]. When the skin phantom is fitted into the array it lies
and the th
antenna
elements. This
distance
backscatter signal20
ofmm
theabove
benign
abnormality
mimic
thatbetween
of theantennas
malignancy.
directly
pointThis
is included
in the DAS algorithm
and breast provides a full coverage of a breast by an antenna antenna point towards center of curvat
radiation
pattern.
direction of a boresight radiation.
affects the detected
signal’s
strength.
The electrical parameters of the skin layer were chosen according to the previously published data. The material is disperV. 3D IMAGING RE
Additionalsive
challenges
associated
with UWB
imaging
and, at 6 GHz,
it has a relative
dielectric
constanttechniques
of 30 and do exist [87]. To
attenuation of 16 dB/cm. Finally, a tumor phantom material with A. Comparison With Previous Antenn
achieve the desired
imagedielectric
resolution,
optimized
of the7 S/m
backscatter
signals is
a relative
constant
close to 50 focusing
and conductivity
In this section we compare 3D imagi
(at 6 GHz) was developed. The contrast between dielectric prop- our previous antenna array [shown in
erties
of breast fat
phantominterface
materials isgenerates
around 1:5.scatter
required. Dielectric
mismatch
at and
the tumor
skin-breast
responses
at 1(b)]. Below we p
developed
array [Fig.
Recently published data on the electromagnetic (EM) properties a 10 mm (diameter) tumor phantom, lo
of breast tissues [11] suggest that the contrast might be signifi. In both cas
least one order ofcantly
magnitude
greater than any tumor response, which is undesirable.
lower, and also that the breast interior is more inhomoge- phantom as described in Section IV.B.
neous than indicated in previously published reports [8]–[10]. tween both breast phantoms used was
Furthermore, theThese
tumor
response
can implications
be masked
by the perforbreast’s heterogeneous
findings
have serious
on achievable
with the 31-elements array had 9 mm
mance of microwave imaging modalities (radar-based, as well
mm, as shown in Fig. 2(a)). F
as inverse scattering).
the skin layer had a radius of
24
C. Differential Imaging and Focusing Algorithm
Before applying the focusing algorithm, the tumor response
must be extracted from measured data. To do so we perform
with the 16-element array was slightly
that the whole array was formed as a p
78 mm, and the radius of the 31-eleme
To quantitatively assess imaging res
composition. Although signal-processing strategies have been developed to reduce the
undesired scatter, they were originally evaluated using numerical breast phantoms with
limited breast heterogeneity. Algorithm advancements may have occurred; however,
when applied clinically their use may be less effective than initially proposed in [87] due
to the combination of attenuation issues known to occur while operating at higher
frequency and the limited SNRs associated with these systems’ designs.
To limit issues associated with optimized beam focusing, Fear et al have
developed a monostatic tissue-sensing-adaptive-radar (TSAR) method (figure 1.8) where
a single balanced antipodal Vivaldi antenna is scanned around the breast perimeter to
collect S11 measurements using a VNA (8722ES, Agilent Technologies, Palo Alto, CA)
[36]. Their single sensor was designed to produce a focused beam that increases the
reflected power, especially for small features. Although still in the experimental phase,
their group has scanned several volunteers using the prototype system, and has indicated
Figure 1.8. University of Calgary’s TSAR system used to scan volunteers.
25
excellent agreement between the measured signals of the volunteer and the corresponding
volunteer-specific MRI based breast model. Top and side views their TSAR system are
shown in figure 1.9.
(a)
(b)
Figure 1.9. (a) Top and (b) side views of Calgary’s TSAR prototype system with
dimension; additional antenna locations are shown (shaded) to illustrate the tank
rotation.
Despite the fact that UWB techniques do not involve numerical inversion
algorithms, and the field’s subsequent desire to highlight the techniques limited
26
computational challenges as a positive attribute, advances in computational power and
the implementation of parallel computing strategies now permit full 3D inversion
tomography. Although the inversion is conducted after the collection of the tomographic
data, additional signal processing is not required. Our decision to utilize a tomographic
approach minimizes typical UWB issues related to measurement signal power, limited
SNRs and the requirement for additional signal processing procedures.
27
1.4.2. Tomographic imaging techniques
Unlike UWB techniques, tomographic approaches can localize dielectrically
contrasting regions within the breast volume and directly recover the tissue’s DPs. By
placing the breast volume in the center of our system’s circular array of monopole
antennas, microwave signals are transmitted through the breast volume and measured
around its perimeter. Calibrated measurement data serves as the input to a post-process
reconstruction inversion algorithm [93, 94, 95], resulting in the generation of the imaged
volume’s spatially dependent DP profile.
A variety of methods have been evaluated for MT system development, each
involving design related trade-offs directly related to the implemented hardware and
software strategy. Possible solutions have been identified and materialized as a result of
these investigations, most of which are still in their experimental phase. Several efforts
have reached the phantom validation stage, but few have promise to successfully translate
to the clinic as a result of design related matters due to limited SNRs, low levels of
channel isolation and antenna related signal issues such as mutual coupling. However,
some of these systems have demonstrated accurate and reliable phantom-based
measurement capabilities, with a limited number successfully expanding to the clinical
trial stage for BC related investigations [34, 41]. Current MT methods under
consideration include UWB time-domain, single and multi-frequency MT systems.
Researchers at Chalmers University of Technology (Gothenburg, Sweden) have
developed an advanced UWB time-domain MT system utilizing an impulse generator
(PSPL 3500D) and dual channel 50 GHz plug-in (HP 54752A) coupled to a digitizingoscilloscope mainframe (HP 54750A) for signal generation and extraction, respectively
28
[40]. Their computer controlled experimental system utilizes an IEEE-4888 bus, allowing
automated measurement acquisition. Figure 1.10 shows a photograph of the imaging
systems’ illumination chamber and 20-monopole antenna array. Their inversion
technique utilizes an iterative gradient algorithm and a finite-difference-time-domain
(FDTD) forward solver, which is fully described in [96]. Promising results have been
reported using this technique [40], especially for the recovered permittivity values.
(a)
Figure 1.10. (a) Photograph of the experimental illumination chamber and antenna
array configuration.
29
respectively, and is used to improve the overall sensitivity of the receiver.
The passive mixer improves the receiver linearity. The mixer (Mini-Circuits A
input signal to the IF band where it receives further amplification and filtering. T
bias and has a conversion loss of typically 8 - 9 dB at the considered frequencies. T
Alternatively, frequency-based tomographic
approaches
haveIFbeen
investigated,
port is amplified
using a two-stage
amplifier
(Analog Devices AD822 [51]) with
The LO signal is routed to the mixer through a network consisting of broadban
the majority of which are being validated. Researchers at Technical University of
the latter of which is used to compensate for the losses caused by the power divisio
Denmark have developed a frequency-basedThe
system
detailed
[34],a total
andgain
have
recently
measurement
units in
provide
of 97
dB with a 2.3 dB noise figur
The noise floor was measured to be below –140 dBm over a 1 kHz bandwidth cent
reported imaging their first patients. Their system (figure 1.11a) utilizes two Agilent
In transmit mode, a harmonic RF signal is fed into the antenna from the signal
N518A-506 signal generators (Agilent Technologies
Palo Alto,
CA)(M/A-COM
to provide
the radio [52]) and the SPST
multiplexed -through
the SPDT
MASW-007071
crowaves, Antennas and Propagation, vol. 4, issue 12, December 2010, pp. 2200–2211
switches. Two cascaded SPST switches are implemented in order to achieve
channel
terodyne
ve
topology
principles,
architecture
in
is
based
which
highly
is
on
operating in receiving mode. The isolation between the channels of the imaging s
an
sensitive
entation receivers. This architecture can
Antenna
System
Computer
Data
Acquisition Power
Supply
System
antennas mounted next to each other, and, as expected, better isolation is achieved
sides of the imaging domain. This value was measured at 900 MHz and it decreas
he dynamic range compared to the more
nly used combination of a two-port
the received fiel
analyzer connected to a switching
to
Antenna
[4], [39] with the dynamic range of the
terodyne imaging system reported in [3]
LNA
ng 120 dB. Also, due to the very low
Measurement
Units
levels in the receiving channels, the
connecting the antennas to the switching
RF Amp.
ze the losses in the cables and to
ze cross-talk and other influences from
Mixer
outside of the individual receiver chain.
plifying and down-converting the signal
t the terminals of the antennas, the
quisition system becomes non-critical
from LO
generator
susceptible to outside noise sources.
multi-channel architecture also allows
fast
data
acquisition,
since
Power
Dividers
the
ements are performed simultaneously for
IF Amp.
from RF
generator
ion from the measurement units and the
Signal
Generators
(a)
(b)
Fig. 2. Photograph of the 3D microwave imaging system.
iving channels. All 32 measurement units are identical and their mode of operation (transmit or receive) is
ed by digital outputs of the data acquisition system connected to the switches in the measurement units. The
ion time for one system state (one antenna transmitting, 31 receiving) is approximately 1.5 second using
the limiting fac
isolation between
Switches
would become critical – both to
50 dB) at the an
Fig. 5. Photograph of the measurement unit with shielding
fences removed. The block diagram of the unit is shown in
Fig. 1.
samples per channel. The choice of samples per channel is a trade-off between the measurement speed and the
1.11.
(a) Photograph
oor of the system since fewer samples perFigure
channel reduces
the measurement
time but also increasesof
the TUD’s MI system and (b) photograph of the
e noise floor of the system. The overall acquisition
time for all states unit
necessary
to create
a 3D image, i.e.,fences
all
measurement
with
shielding
removed.
nnas, in turn, acting as transmitters, is approximately 50 seconds, including the necessary switching time for
d-state SPDT switches as well as time needed for storing the measured data on the computer.
following subsections are dedicated to a more detailed description of custom designed components of the
frequency (RF) and local oscillator (LO) signals to the system’s 32 measurement units
(figure 11b) using two 32-way power division (PD) networks (one of the RF and one for
the LO). These networks consist of one 2-way PD in-line with two 16-way broadband
Wilkinson PDs. A modulated intermediate frequency (IF) signal is digitized by a 32
30
side of the imagin
The measurem
supply. A view
shielding fences
built on Rogers R
of the module (
circuit distributio
circuit board, esp
signals, guarante
operation, which
channel 18-bit analog-to-digital (A/D) converter, and processed using a LabVIEW signalprocessing program. Their initial results are promising; however inherent isolation issues
may exist due to the structure of the RF and LO networks.
Dartmouth’s Microwave Imaging Spectroscopy Group has successfully evaluated
and built a frequency-based MT system that is currently utilized in clinical trials at
Dartmouth Hitchcock Medical Center (Lebanon, NH). To the best of the author’s
knowledge, our system currently represents the most advanced clinical MT system, and
has been investigated for applications related to BC detection and diagnosis [88, 97],
chemotherapy monitoring [89] and hyperthermia [98]. A thorough description of the
system’s design and performance will be discussed in Chapter 2 and Chapter 3,
respectively.
The decision to use an Agilent ESG-D series 4432B synthesized RF signal
generator (Agilent Technologies Palo Alto, CA) as our system’s signal source reduces the
SNR issues associated with VNA time-domain techniques. Additionally, its design
incorporates a switching matrix and a PD network to supply the transceiver units’ RF and
LO signals, respectively. We believe the utilization of a dedicated RF switching matrix
allows optimal performance when compared to a coupled PD based RF network, which
may induce EM coupling between circuit elements [Z2].
Furthermore, our chosen antenna design and coupling medium are well suited for
BC applications. The antennae can be placed close to the breast surface in high density
due to their small size. Moreover, the lossy coupling medium acts as a matching layer
reducing skin surface scattering without the need for additional signal processing; it
attenuates undesired scatter signal and can be tailored to a variety of breast densities. Due
31
their combined use, the acquisition of reliable and clinically relevant data with limited
signal corruption issues has been demonstrated [88, 97].
Despite UWB’s potential clinical applicability to BC detection and diagnosis,
issues related to time-domain scattering phenomena, broadband antenna design, and
volumetric beam coverage are still prominent. These issues, however, do not affect the
integrity of our frequency-based MT system’s measurements. Undesired scattered signals
and additional noise issues are minimized due to our system’s calibration scheme, and
with recent advances to the reconstruction algorithm [99], full 3D volumetric
tomographic MI has been realized. As a result, the work in this thesis will focus on a 3rd
generation frequency-based MT system, highlighting microwave electronic, data
acquisition and motion control system advances that have resulted in faster data
acquisition speed, increased measurement capacity, improved DP recovery and increased
clinical applicability.
32
1.5 Summary of proposed work
The remaining chapters will analyze the following system developments:
1) Analyze the system’s microwave electronics in terms of channel isolation, system
sensitivity and measurement repeatability in an effort to define an optimal operational
bandwidth (Chapter 2 - 3).
2) Develop a signal optimized, user-friendly motion-control system to be used in a
prototype 3D microwave imaging system, and perform a signal sensitivity analysis based
on the separation of the transmitting (TX) and the out-of-plane receiving (RX) antenna
elements. Results from these experiments will allow the determination of positionoptimized 3D motion-control sequences by eliminating the need to acquire data that falls
below the noise floor (NF) of the system’s receiving channels, increasing patient
examination speed and image reconstruction accuracy (Chapter 4 - 5).
3) Facilitate the physical mating of the new imaging hardware and mounted surface
capture system with the current microwave electronics, and develop the software that is
required for communication between the master and slave computer systems. The
outcome of this work has resulted in the materialization of a motion optimized MT
system (Chapter 6).
33
4) Advance the conformal meshing technique for the purpose of generating twodimensional (2D) and 3D conformal microwave images of objects using boundary
information obtained from geometric measurements, magnetic resonant (MR) images and
the optical scanner. Results from these investigations show the enhancement of
dielectrically contrasting regions that reside within both geometrically and irregularly
shaped boundaries (Chapter 7).
34
2. Microwave Tomographic Imaging: A System Overview
The previous generation imaging system (2nd generation) [41] was superior when
compared to its predecessor system (1st generation) [100] in terms of operational
bandwidth and the number of independent measurement sites. The increased operational
bandwidth was intended to provide increased spatial resolution as well as increased
property information for breast cancer diagnosis. Our current (3rd generation) system
continues to use this operational range (500 MHz – 3 GHz) as a result of available
components and antenna design. To increase the number of measurements, the 2nd
generation system sampled all non-transmitting antennae during signal acquisition, where
previously only antennae opposite to the transmitting element were used. The technique
of sampling all non-transmitting antennae has been retained in the current system.
Additionally, a parallel detection scheme was implemented in the 2nd generation
system that allowed the sampling of all measurement sites simultaneous with the use of
two 8-channel 16-bit analog–to-digital (A/D) boards and a programmable gain amplifier
[41]. The limited dynamic range of the 16-bit A/D boards necessitated additional signal
amplification to account for the increased signal attenuation that is known to occur while
operating at higher frequencies.
The 1st generation system utilized a single signal
demodulator and single A/D board to iteratively sample each receiving element
sequentially [100]. The parallel detection scheme dramatically increased the dataacquisition speed of the 2nd generation system, and has been re-implemented in the
current design.
In the 3rd generation system, key data-acquisition elements have been replaced by
newer components with larger dynamic ranges: the two 8-channel 16-bit A/D boards (NI
35
SCXI-1125) and programmable gain amplifier combination used in the 2nd generation
system have been replaced with two 8-channel 24-bit A/D boards (NI PXI-4472). With a
sampling rate of 102.4 kS , these A/D boards are capable of covering the entire
dynamic range without additional signal amplification. This advancement has increased
data acquisition speed by eliminating the dynamic amplification procedure.
In an effort to reduce the system’s overall size, alterations made to the RF and LO
networks have reduced its footprint by replacing the 2nd generation’s 32-channel design
to 16. These modifications have streamlined the configured layout by eliminating
redundant componentry while attempting to maintain a desired level of channel isolation,
allowing the development of a compact and portable, self-contained, clinical MI system.
The changes to the A/D boards and the RF and LO networks have altered the current
system’s microwave electronic performance, and will be analyzed in Chapter 3.
The new system is modular in design. Electrical components have been
compartmentalized, and are physically separated from the patient interface and aqueous
coupling medium; increasing patient safety. This has also increased the system’s potential
as a clinical tool. Connection of the patient interface and imaging antenna array to the
electronic system is made through a bulkhead panel with mounted SMA connectors. This
feature facilitates the attachment of various imaging arrays, permitting multimodal MI
applications that were previously unfeasible. For example, the connection of MR
compatible MT imaging arrays to the system’s electronics permits operation in strong
magnetic fields, such as a magnetic resonant imaging (MRI) system’s bore [101]. We
have again exploited this attribute for the connection of a newly constructed patient-
36
imaging interface that allows optimized data acquisition and the extraction of the breast’s
boundary while submerged in the system’s imaging tank.
37
2.1 System hardware layout
Dartmouth College’s new (3rd generation) MT imaging system utilizes a circular
monopole antenna array (diameter = 15.2 cm) housed in an associated illumination
chamber (figure 2.1a). The 16-element array is composed of two interleaved 8-element
sub-arrays (SA), with alternating antenna elements belonging to complementary SA sets
(figure 2.1b). Each SA is rigidly attached to a separate vertically translating mountingplate (figure 2.1c). This SA configuration and the associated mounting-plate attachment
(a)
(a)
(b)
(d)
(c)
(c)
(e)
(b)
(d)
Figure 2.1. Photographs and computer renderings of Dartmouth’s MT imaging system
1.2. MIST
system:housed
(a) antenna
configurationillumination
on two independently
independently-moving
moving plates A
highlighting (a) 16-elementFigure
monopole
antenna
in associated
chamber,
(b) schematic of the two (pink)
interwoven
SAs,
(c) SAtank,
attachment
to independent
and B (blue),
(b) imaging
((c) exam platform,
and (d)) cablinglinear
and fluid reservoir
actuator-controlled mounting-plates, (d) SAs in an in-plane configuration and (e) SAs in
underneath the table
an out-of-plane configuration.
38
allows independent motion-control of the array subsets such that the two can move to a
number of in-plane and out-of-plane positions (figure 2.1d and 2.1e). It is this
independent SA control that enables our system to transmit and receive volumetric
microwave signals.
Figure 2.2. System connectivity diagram highlighting the RF, LO and antenna inputs to the
transceiver units, as well as the system’s parallelization.
A diagram of the MI system’s hardware layout is shown in figure 2.2. Antenna
elements can operate as both a transmitter and a receiver with the use of individually
dedicated transceiver units. These 16-independent antenna-transceiver connections form
39
the channels utilized in the system’s parallel detection scheme [41]. A LO signal is fed
through a series of PDs, producing coherent reference signals for each channel of the
parallel detection design. The detected microwave signals are mixed with the LO signal,
and the demodulated, low pass filtered IF signals are detected in parallel by the system’s
data acquisition system (DAQ). A software based lock-in amplifier technique is used to
extract amplitude and phase values from the detected microwave signals simultaneously.
For our system to successfully detect useful transmission data, only one antenna
can be radiating at a time. All remaining, non-transmitting antennae act as signal
receivers. Although each of the 16-antennae acts as a transmitter during the data
acquisition process, our approach is sequential; while a single antenna is transmitting
every other array elements is receiving. This approach is accomplished with the use of a
bi-level switching matrix (section 2.2.2); the RF signal is fed directly to the desired
transmission element while the remaining elements detect the transmission data in
parallel.
After data from the first transmitter has been detected by all complementary array
elements, the switching matrix guides the transmission signal to the next antenna. By
using this switch-based iterative approach, each of the 16 antennas independently and
asynchronously act as a transmitter while the complementary antenna elements
simultaneously act as receivers. Data acquisition for a single frequency acquired by all
antennae in a single plane results in 240 independent measurements.
40
2.2 Electronic hardware design
Our system’s reconstruction algorithm requires the use of the real and imaginary
parts of the detected complex microwave signal from every receiving site. A signal
heterodyning technique, utilized for amplitude and phase measurement, was successfully
employed in the 1st and 2nd generation systems, and has been re-implemented in the
current design. Key elements such as the PDs and the A/D boards have been replaced by
newer, cost-effective commercially available components with increased operational
bandwidths. This upgrade was intended to improve system response over a wider
frequency range while increasing its data acquisition speeds. Changes to the RF and LO
network layouts have compacted the system’s overall size, allowing the construction of
an enclosed modularized system, where the electronics, DAQ components, power
supplies and liquid-coupled patient interface have been compartmentalized (figure 2.3a).
The massage table-like top (figure 2.3b) has an aperture that facilitates comfortable
1
3
2
4
(a)
(b)
Figure 2.3. (a) Photograph of Dartmouth’s MT system highlighting its modular
design including 1-illumination chamber, 2-microwave electronics’ compartment, 3DAQ components, 4-power supply arrangement and (b) photograph of the system
covered with customized patient table with breast aperture.
41
placement of the patient’s breast into the imaging chamber while resting prone.
Like the predecessor system, the current design consists of three modules: (1)
microwave electronics (sections 2.2.1-2.2.3), (2) liquid-coupled patient interface (section
2.3) and (3) computer control and data acquisition (section 2.4), each of which will be
discussed in this chapter.
42
2.2.1. Microwave source and LO network
An Agilent ESG-D series 4432B digital RF signal generator (rfSG) serves as the
system’s microwave signal source. It offers customizable test functions; nearly every
aspect of the digital signal or signal operating environment can be adjusted, which is
advantageous for nonstandard wireless protocols [102]. The rfSG has a frequency range
of 250 kHz to 3 GHz, a calibrated variable power output ranging from +10 to -136 dBm
and an adjustable IF. Compared to the previous system’s signal source, which yielded a
sideband and carrier compression down to
60 dBc over a range of 250 kHz to 3 GHz,
the current rfSG yields a modulated signal with sideband and carrier compression down
to -80 dBc over the same frequency range. Additionally, LO and
signal outputs are
located on the backside of the generator for signal demodulation and lock-in amplifier
based and
recovery, where and
represent the in-phase and quadrature components
of the detected microwave signals, respectively. With the use of a GPIB data bus, the
rfSG can be integrated into the automated computer-controlled system.
The LO signal from the rfSG is fed to the PD network consisting of one 2-way PD
(PS2-22-450/10S Pulsar Microwave Corporation, Clifton, NJ) and two 8-way PDs (PS812-454/50S Pulsar Microwave Corporation, Clifton, NJ), resulting in 16 coherent
reference signals, one for each channel of the parallel detection scheme. The PD’s
frequency range is .5 – 6 GHz with a maximum VSWR of 1.40. Over our system’s
operational frequency range (500 MHz – 3 GHz), their insertion losses are less than 0.6
and 1.5 dB for the 2-way and 8-way PD, respectively. Additionally, their minimum
output channel isolation is 18 dB and 20 dB for the 2-way and 8-way, respectively [103].
43
As with the 2nd generation system, coaxial attenuators have been placed between
the PD stages to suppress insertion loss mismatch effects. An amplification stage has
been configured into network (M/A-COM MAAM020350-A2, Lowell, MA) to boost the
PD’s output signal strength to the level required for powering the transceiver module’s
mixer (Minicircuits ADE-30). The mixer’s power requirements vary as a function of
frequency, and unfortunately the rfSG’s LO signal is output at a constant power level. As
a result, a transistor-transistor logic (TTL)-controlled digital attenuator (M/A-COM AT213) has been introduced into the network since a fixed attenuation level did not meet the
mixer’s varying drive power requirements over the full frequency range. The input power
levels have been optimized in an effort to minimize the mixer’s conversion loss without
increasing higher order harmonics. A schematic of the PD network is shown in figure 2.4,
and the corresponding computer aided design (CAD) drawings of the 2-way and 8-way
PDs are show in figure 2.5.
Figure 2.4. Diagram of new LO PD network highlighting the streamlined design
configuration.
44
(a)
(b)
Figure 2.5. CAD drawings of the (a) 2-way and (b) 8-way PDs (compliments of Pulsar
Microwave Corporation).
45
2.2.2. Transmitting/receiving channels
Due to the system’s iterative signal transmission strategy and its parallel detection
scheme, fast switching and well-matched electronics with high inter-channel isolation are
required for signal transmission and detection. To allow each of the 16 channels and their
associated antennas to independently and sequentially transmit microwave signals, a bilevel 16-channel switching matrix, composed of 5 broadband single-pole-four-throw
(SP4T) switches (M/A-COM SW311) was designed. These switches have inter-channel
isolation greater than 40 dB over our system’s frequency bandwidth. Figure 2.6 shows a
schematic of the switching matrix, along with a photo of an individual SP4T switch.
SP4T
Input
16 channels
SP4T
SP4T
SP4T
SP4T
(a)
(b)
Figure 2.6. Diagram of the new RF switching matrix highlighting the bi-level SP4T
switch configuration and (b) a photograph of a single SP4T switch.
46
The transceiver unit is a fundamental component of our MI system, and is
required for system operation; antenna elements in the imaging array can operate as both
a transmitter (TX) and a receiver (RX) as a result of the independent, channel dedicated
transceiver units. As shown in figure 2.7, the transceiver unit includes 3 amplifier
chambers, a series of internally matched single-pole-single-throw (SPST) and singlepole-double-throw (SPDT) switches (M/A-COM SW311 & SW313), and a mixer
(Minicircuits ADE-30). The cascade of the SPDT and SPST switches allow the
transceiver to operate in both a TX and RX mode. Although the SPDT switch alone can
facilitate the bi-modal characteristic, the incorporation of the SPST switch has increase
network and channel isolation.
Figure 2.7. A schematic of the system’s transceiver unit diagraming the layout of its
electronic components, including amplifiers, switches, and a mixer; also shown is the RF,
LO, Antenna Port and A/D board connection points.
47
Two amplifiers (M/A-COM MAAM020350-A2), one positioned after the RF input and
the other positioned between the LO input and the mixer, compensate for insertion losses
from the switching matrix and the LO network, respectively. The additional amplifiers
(M/A-COM MAAM020350-A2) located between the SPDT switch and the mixerchamber strengthens the received microwave signals, which are highly attenuated due to
the lossy coupling medium and breast tissue. By inserting these low-noise amplifiers near
the beginning of the receiver’s component cascade, we can maintain a low system noise
figure [41, 103]. Additionally, a coplanar waveguide was used as the transmission
medium in the circuit layout to minimize transmission loss and the radiation of spurious
signal modes as a result of the mixer’s 6.6 dB conversion loss and the LO drive power of
+7 to +10 dBm.
48
2.2.3. System integration
The mounting scheme is also modular in design; components of the PD network
and the switching matrix have been attached to separate square mounting plates, and the
transceiver units are mounted on the electronics’ compartment wall, right behind the LO
and switch matrix plates (figure 2.8a). This design allows easy attachment of the
transceiver units to the antenna elements through a bulkhead panel positioned in the wall
1
2
3
(a)
(b)
(c)
(d)
Figure 2.8. Photographs of (a) mounted modular assembly of 1- RF switch matrix
module, 2- LO PD module and 3- Wall mounted transceiver unit assembly, (b)
Bulkhead mounting plate used for transceiver-antenna connection as seen from the
microwave electronics’ compartment, (c) Bulkhead mounting plate used for transceiverantenna connection as seen from the illumination chamber cabinet, and (d) Mounting
module terminal block connection.
49
that separates the electronic and imaging compartments (figures 2.8b and 2.8c). The close
proximity of the PD and switch matrix mounting plates to the wall mounted transceiver
units facilitates their connection using short coaxial cables, limiting potential cable
related attenuation issues using longer transmission lines. Each mounting module
includes a terminal block for power cable and component control line attachment (figure
2.8d). Two direct current (DC) power supplies (E3631A) positioned in an adjacent
compartment containing the DAQ chassis and rfSG power the entire electronics system
(figure 2.9).
1
2
3
(a)
(b)
Figure 2.9. Photographs of (a) commercial electronic hardware cabinet including 1DAQ chassis, 2- DC power supplies, 3- RF signal generator and (b) positioning of
microwave and non-microwave electronics’ cabinets.
50
2.3 Illumination chamber and liquid-coupled clinical interface
The illumination chamber houses an array of 16 monopole antenna elements
immersed in a glycerin/water-coupling medium (figure 2.10). The coupling medium
(shown in its storage reservoir) acts as a matching layer, limiting reflections at the breast
surface in an effort to increase scattering within the breast volume. Additionally, the
lossy nature of the coupling medium resistively loads the antenna elements, broadening
their frequency response; this phenomenon, along with an analysis of the system’s
antenna design will be discussed in section 2.3.1. The cylindrical design of the
illumination chamber serves to uniformly randomize chamber related signal scattering,
although the signals are presumably highly attenuated as a result of traveling through the
lossy medium.
1
2
3
Figure 2.10. Photograph of the 3rd generation system’s clinical interface including 1illumination chamber, 2- antenna array and 3- coupling medium.
51
Four stepper motor (Parker Hannifin (PH) S57-83, Cleveland, OH) driven electric
cylinder linear actuators (PH ET32 803-4960A-269) symmetrically located underneath
the illumination chamber, attach to two independent SA mounting plates. The motors are
powered by an Omron 24V DC power source (PS-S8VS-1202B) located in the power
control cabinet. The linear actuators are coupled, such that each pair is attached to and
controls a separate mounting plate. The 16 element antenna array is composed of two 8
element SAs; neighboring antenna elements are associated with the complementary SA.
This dual mounting plate control is an important design innovation. The ability to
independently move the antenna SA to various in-plane and cross-plane locations within
the tank has increased the number of independent measurement sites, facilitating 3D
volumetric data acquisition.
(a)
(b)
Figure 2.11. Photograph of the antennas (a) entering the illumination tank from the
bottom of the imaging apparatus through the base plate holes and seal-related O-rings
and (b) a top-down view of the antenna array inside the tank with Teflon seals in
place.
52
Antennas are attached to their corresponding SA, and enter the illumination
chamber through holes in the base of the tank (figure 2.11). This design keeps all
electrical connections outside and underneath the illumination chamber, insuring a
physical separation between the coupling-medium and the antenna’s electrical
connections. To prevent leakage through the base plate antenna holes, custom designed
seals consisting of a Teflon outer shell and two internal O-rings are configured into the
base of the tank (figure 2.11). The seal has a hole in its center that allows the antenna to
enter the tank; screwing the seals into the base plate compresses the O-rings, creating a
liquid tight closure. The linear actuator controlled mounting plates allow vertical
movement of the antennas inside the illumination chamber.
Vertical translation of 12.5 cm is possible; although based on the author’s
experience 7 cm of vertical movement is sufficient to image the entire breast volume for
most patients. When the breast is submerged in the illumination tank’s coupling medium
it tends to expand horizontally and compress vertically due to buoyancy since the
coupling medium is more dense than the breast. A micro-controller is used to regulate the
motors movement; each motor-actuator complex can generate 135 f-lb of torque through
a 1:1 gear ratio. With a resolution of 0.0254 mm at 25,000 steps per revolution, a
maximum velocity of 396 mm/s can be obtained. The high torque level achieved by the
four motor assemblies is required to drive the actuators with sufficient force to overcome
the compressed O-ring seals. The micro-controller is connected to a computer that
controls the entire system, allowing user control of motion parameters such as speed,
force and position, although their speed and force are rarely changed.
53
Additionally, the current system incorporated a DC isolated pump that is
controlled through the GUI. A level sensor in the top of the tank automatically stops the
pump when the tank is full. This is a beneficial design implementation; the requisite
coupling medium level can be achieved automatically, regardless of the size of the
patient’s breast volume. Also, fine adjustments can be made to the fluid’s level remotely,
which is important when a patient’s breasts are not the same size. The filling and
draining of the tank is accomplished using a series of hydraulic valves, and due to the its
variable speed control, filing and draining times can be optimized allowing fast and easy
cleaning of the system after the patient exam has ended. A picture of the pump and fluid
valve design are shown in figure 2.12.
(a)
(b)
Figure 2.12. Photographs of (a) the pump used for filling and draining the tank and
(b) the valve connection from the coupling mediums storage container to the imaging
tank.
54
2.3.1. Antenna design and characterization
For an imaging system to be clinically useful it must provide reliable data; this
requires the detection and extraction of accurate microwave signal information. Due to
their emission and detection responsibilities, the antenna is a fundamental component of
every MI system, and can heavily affect the integrity of its operational performance.
Taking the varying frequency bandwidth requirements of the implemented design
strategy (UWB or tomography) into account, proper antenna selection is a primary
determinant of a system’s measurement capabilities. And consequently, the system’s
measurement accuracy is directly related to the characteristics and performance of the
utilized antenna design.
Four antenna types have been used and identified as potential candidates for
microwave BC imaging applications; namely: the monopole antenna [104], the bow-tie
(a)
(b)
Figure 2.13. (a) CAD drawing of our system’s monopole antenna (compliments of
Dr. Tain Zhou) and (b) photograph the two monopole antennas utilized in our dual SA
design (also shown is one monopole with its active part exposed).
55
antenna [105, 106], the Vivaldi antenna [107, 108, 109] and the pyramidal horn antenna
[87, 110, 111, 112]. Each of these antennae are associated with varying design-dependent
performance characteristics. Our system utilizes in-house constructed monopole antennas
(figure 2.13), assembled with a 1/16 diameter stainless steel (SS303) inner conducting
rod, a center dielectric made of Teflon (
2.1, 1/16 ID and 3/16 OD) and a stainless
steel (SS303) outer conductor (2/11 ID and 5/16 OD). Associated design dependent
characteristics include an impedance of 45.4
, a reflection coefficient of 0.048 and a
voltage standing wave ratio (VSWR) of 1.1 [82], where a VSWR of 1 implies a matched
load [102]. The length of the antenna’s active part is 3.4 cm; chosen due to its low return
loss over our system’s operational frequency band (-22 dB at its resonant frequency of 1
GHz).
In free space the monopole has a narrow bandwidth, a circularly symmetric beam
pattern and is known to excite surface currents along its outer conductor [102, 113, 114].
However, operation in the system’s lossy coupling medium broadens the antenna’s
frequency response, presumably due to resistive loading. The attenuating nature of the
medium limits surface current excitation, and despite the radiation pattern, near field
performance within the lossy media has resulted in image quality enhancement [115].
This monopole design has been advantageous for our multichannel motion
enabled array system: it is easily modeled [116] and can be positioned close to the breast
surface. High element density can be achieved when the elements are arranged in an
imaging array due to low mutual coupling, and their construction process is relatively
easy and low in cost. A thorough characterization of the system’s antenna elements in
56
terms of return loss, effective beam width and beam steering angle as a function of
electrical length can be found in [104].
57
2.4 Computer control and data extraction
The entire system is automated and controlled by a single personal computer
(PC) through a PXI Embedded Controller (NI PXI-8176, National Instruments (NI)
Austin, TX) utilizing a GPIB Ethernet Controller (NI PXI-8231 NI, Austin TX)
connection. The rfSG and the power supplies are controlled through 8-Slot 3U Chassis
with Universal Power Supply (NI PXI-1042 NI, Austin, TX) (figure 2.14). Array motion
is controlled through a RS232 serial port in conjunction with the micro-controller. Logic
control of the switching matrix and digital attenuator is accomplished using a Digital I/O
Module (NI PXI-6508 NI, Austin, TX).
Signal measurements are mixed with the LO signal and the modulated, low-pass
filtered IF output is sampled in parallel by a two 8-Channel 24-bit Dynamic Signal
Acquisition (NI PXI-4472, NI, Austin, TX) boards at a sampling rate of 102.4
.
Amplitude and phase measurements are extracted from the IF signal using a software-
Figure 2.14. Photograph of the system’s DAQ chassis showing Ethernet connection,
digital I/O connection and IF signal inputs to the two 8-channel 24-bit dynamic signal
acquisition boards.
58
based lock-in amplifier technique, and serve as inputs to the reconstruction algorithm.
Using the 24-bit A/D boards has increased our system’s data acquisition speed when
compared to the 2nd generation system, which incorporated an additional dynamic
amplification stage due to the limited dynamic range of the 16-bit A/D board. The
acquisition time for a complete set of data (240 measurements) at a single frequency is
5.8 seconds, roughly 3 seconds faster than the previous system. This is clinically
advantageous; faster data acquisition speeds decrease patient exam times, allowing the
collection of more data than the previous system over the same period of time. Overall
system operation, including signal conditioning, A/D conversion, the logical controlling
of switching matrices and antenna motion coordination is controlled with custom
software written in LabVIEW (NI). A customized, user-friendly graphical user interface
(GUI) allows system operation from a PC located outside of the patient examination
room.
59
2.5 Homogeneous data calibration
To accurately assess the DP of any object under test (OUT), a calibration process
is utilized to remove the unwanted 3D free-space loss factors [115] as well as system
related noise. This is accomplished by subtracting measurement values acquired with a
homogeneous medium {
presence of the OUT {
} from the measurement values acquired in the
} under the same data acquisition conditions.
Using this calibration process, field variations specifically related to the OUT, {
},
are given by:
{
}
{
}
{
}
(2.1)
The resulting calibrated signal amplitude and phase measurements are used as inputs to a
post-processing reconstruction algorithm [93, 94, 95].
60
2.6 Image Reconstruction
Based on the limited resolution associated with microwave diffraction
tomography, we have implemented a computational scheme to estimate the DP values of
the OUT. Due to the ill-posed and non-linear nature of solving the wave equation for the
squared complex-value wave number, given as:
( )
Where
=(
( )
)
( )
(2.2)
and c is the speed of light in the material, we have implemented
a Gauss-Newton (GN) iterative technique with a log-transformation [94], Tikhonov
regularization [93] and a FDTD forward solver to reconstruct the DPs of the imaged
volume [95]. The method seeks to minimize the differences between the measured and
calculated amplitude and phase values using a least squares approach. Additionally, the
use of a dual-mesh approach [117] has been implemented; the forward solution is
calculated on a FDTD grid and the DP parameters are reconstructed on a triangular (2D)
or tetrahedral (3D) mesh.
The objective function for our algorithm is represented as:
‖
(
)‖
‖
(
61
)‖
‖ (
‖
(2.3)
where
and
are the log magnitudes and
and computed field values, respectively.
the regularization matrix,
conductivity) with
and
are the phases of the measured
is the Tikhonov regularization parameter,
is
includes the property estimates (both permittivity and
representing its prior estimate. In our uniform reconstruction
approach, the regularization matrix is set equal to the identity matrix, which applies the
same weighting to all nodes in the imaging domain. During the iterative reconstruction
process, the updating of
is based on data-model misfit; the updating procedure is
given by:
{
where {
}
{
}
(2.4)
} is formed from:
{
}
{
}
{
where the Jacobian, , is defined as [
}
(
)
(
)
(2.5)
], i is the iteration number, and
and
the measured and computed electric fields, respectively. A least square fit for {
are
} is
obtained by solving the set of normal equations constructed by multiplying both sides by
of equation 2.4 by
such that:
{
}
{
}
62
(2.6)
Once the updated parameter has been calculated, a new forward solution is computed.
The iterations continue until the data-model misfit is less than a predetermined level. Our
3D inversion model has implemented a parallel computing strategy where the forward
solutions for each antenna are calculated simultaneously [99].
63
2.7 Conclusion
This chapter has outlined the current imaging system’s microwave electronic
networks, signal transmission componentry and illumination chamber components.
Compared to its predecessor, the current system is more compact and modular.
Modularity has increased patient safety by physically separating electrical components
from the patient interface while increasing its functionality. It permits easy removal and
replacement of the illumination chamber with a variety of antenna arrays; consequently
the current system can operate within high strength magnetic fields, enabling multi-modal
MT imaging. The use of the monopole antenna has been advantageous for our
application; although not optimal in free space, operation within the lossy coupling
medium allows high-density placement of the antennae close to the breast surface while
broadening their operation frequency range. The system has shown clinical relevance for
breast cancer detection, diagnosis and chemotherapy monitoring.
64
3. Electronic Hardware Performance
Reconfiguring the microwave electronics’, DAQ components and motion system
has increased the system’s data acquisition speed and measurement capabilities; clinically
relevant advances allowing faster imaging procedures while collecting more patient data
for improved image reconstruction. To ensure the accuracy and efficacy of the updated
system’s measurement abilities, a thorough analysis of the electronics’ performance has
been conducted. This chapter will evaluate its operational characteristics, in terms of
system sensitivity, channel isolation and measurement repeatability in an effort to define
an optimal operational frequency range. A schematic of the new system is shown in
figure 3.1.
Figure 3.1. Schematic of the imaging system highlighting the incorporation of the new RF
and LO networks (only shown for four transceiver connections). Output and measurement
signal paths are shown in blue and black, respectively.
65
3.1 System sensitivity and error
The power of the lowest detectable signal that a system can measure is easily
converted to its minimum detectable voltage signal; this is commonly referred to as
receiver voltage sensitivity, or just receiver sensitivity [102]. To evaluate the system’s
sensitivity and error, each of its channels was characterized using the Agilent rfSG. One
goal in the redesign of the system was to maintain linear output as a function of the input
power. For this analysis, each channel was directly connected to the RF source with a 10
dB attenuator placed in-line. The rfSG has a linear variable output range of +10 to -136
dBm. As a result, signal phase and amplitude measurements were acquired over a wide
range of calibrated input power levels. Using the system’s DAQ, amplitude and phase
measurements from 500 to 2900 MHz in 300 MHz increments were collected to assess
the sensitivity of a channel’s output to changes in received input power levels. The
resulting measurements are shown in figure 3.2.
Similar to the previous system, phase measurement errors are the limiting factor
in the current system’s ability to detect reliable low power signals. From figure 3.2b it
can be seen that the lowest detectable signal with phase error less than 10 is roughly -130
dBm at frequencies below 2 GHz. Over the system’s entire operational frequency band
this degraded to -110 dBm, presumably due to transmission losses while operating at
higher frequencies, negative gain slope of the broadband amplifiers and diminished
performance of the mixers due to high frequency conversion loss degradation. These
issues directly impacted the resulting high frequency measurements.
Overall, the amplitude [dB] and phase errors increased from average values of
0.325% and less than 0.50 at higher input power levels, to greater than 1% and 1o as the
66
input signal’s power decreased towards the noise floor, respectively. The modifications to
the system’s microwave electronics have increased its sensitivity when compared to the
previous generation system in both amplitude response and phase error. Presumably this
is due to the reduced number of network components, which limits the potential for
component-characteristic related mismatch issues.
Amplitude Sensitivity Response
0
Output Signal Amplitude [dBm]
-20
-40
500 MHz
800 MHz
1100 MHz
1400 MHz
1700 MHz
2000 MHz
2300 MHz
2600 MHz
2900 MHz
-60
-80
-100
-120
-130
-120
-110
-100
-90
-80
Input power [dBm]
-70
-60
-50
-40
-30
(a)
Phase Sensitivity Response
5
4.5
Phase RMS error [degrees]
4
500 MHz
800 MHz
1100 MHz
1400 MHz
1700 MHz
2000 MHz
2300 MHz
2600 MHz
2900 MHz
3.5
3
2.5
2
1.5
1
0.5
0
-130
-120
-110
-100
-90
-80
Input power [dBm]
-70
-60
-50
-40
-30
(b)
Figure 3.2. System (a) amplitude sensitivity response and (b) phase error as a function
of variable input power levels.
67
3.2 Inter-channel isolation
Inter-channel isolation and the resulting coupling of signals between various
pathways in the system were assessed to evaluated possible sources of signal leakage. For
this analysis the antenna port of transceiver #1 was directly connected to the antenna port
of transceiver port #13 using a short coaxial cable. The antenna ports of all remaining
transceiver unit were terminated with 50
matched loads. With channel #1 active, this
set-up directly routed the transmitted signals to channel #13 (relative receiving channel
12), and using the system’s DAQ signal measurements were made at all non-transmitting
channels simultaneously.
We have taken extra care to isolate any potential ground path issues in the
electronics’ powering and feedline configuration. 0.1 f capacitors have been placed on
each channel of the back-end IF amplifier to eliminate high-frequency signal coupling;
future work can focus on reconfiguring the amplifier’s power and ground-lines into a star
formation, which can limit additional low-frequency channel cross-talk. Consequently,
the only remaining leakage paths in the system’s microwave electronics’ design are
through the RF and LO networks.
Figure 3.3. Schematic of the new bi-level RF switching matrix highlighting its two level
design and the connection of channels to their associated SP4T switch (shown for
channels 1 to 4).
68
Results from this analysis revealed interesting design modification-related
performance changes to channels that are connected to the same SP4T switch in the RF
network (shown in figure 3.3 for channels 1 - 4). Elevated signal levels were detected at
all non-transmitting channels of an active switch (figure 3.4.), especially above 1300
MHz. However, when compare to the previous system, which detected signals 15 dB
above the NF at channels adjacent to the receiver (relative receiving channel #11), the
current design reduced this coupling by roughly 5 dB, presumably due to the increased
channel isolation of the 8-way PD (20 dB) compared to the 2-way PD (5 dB).
0
Inter-channel Isolation
500 MHz
900 MHz
1300 MHz
1700 MHz
2100 MHz
2500 MHz
2900 MHz
Received Signal Amplitude [dB]
-20
-40
-60
-80
-100
-120
-140
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Relative Receiving Channel #
Figure 3.4. Plot of measured signal levels (normalized to the desired signal) with channel
#1 directly connected to channel #13 (relative receiver #12) and all other channels
terminated with 50 loads.
69
Although the author assumed the source of the leakage signal was related to
limited isolation in the RF network, a simple experiment was conducted to confirm this
hypothesis. In this investigation, a spectrum analyzer (Agilent ESA-L Series) was
directly connected to the IF output of Channel #2 (which was set to receive mode) while
Channel #1 was actively receiving RF signals. All transceiver antenna port connections
were terminated with 50
matched loads, including Channel #1. By terminating all the
antenna connections, no signals could be radiated, and consequently, no receiver
network-based leakage signals could be introduced into the electronics’ network. This
eliminated any potential antenna-LO based leakages, allowing a direct analysis of signals
specifically related to the RF network. Figure 3.5 and 3.6 show a schematic and
photograph of the experimental set-up used to measure the IF output signal.
Figure 3.5. Schematic of the experimental set-up used to measure the IF signal output
from transceiver unit #2 (LO connections are not displayed and antenna ports are shown
terminated with matched loads).
70
Figure 3.6. Photograph of the experimental set-up shown in figure 3.5 highlighting the
detected leakage signal at transceiver unit #2’s IF port.
To verify that the source was in fact related to the RF switching network,
transceiver unit #2’s RF connection was removed and replaced with a 50
matched load.
A schematic of the modified experimental set-up is shown in figure 3.7; a photograph of
the modified set-up and measured IF signal are shown in figure 3.8. By removing the RF
connection, the leakage signal detected at Channel #2 was immediately eliminated. This
Figure 3.7. Schematic of the modified experimental set-up used to measure the IF
signal output from transceiver unit #2 (LO connections are not displayed and antenna
ports are shown terminated with matched loads).
71
simple experiment proved that the origin of the leakage signal was RF network based.
Determining the origin of this signal was critical, as any future electronic modifications
need to account for RF switch based isolation issues.
Figure 3.8. Photograph of modified experimental set-up shown in figure 3.7
highlighting the modified signal output at transceiver unit #2’s IF port.
In modifying the size of the switching network, we reduced the number of matrix
layers from three to two. Although the previous system was designed with 32 channels,
only 16 were used in the final design. The extra switches provided additional isolation;
with each SP4T switch only utilizing two of its four channels, the resulting 16 channels
experienced enhanced inter-channel isolation due to the network’s RF configuration. S21
measurements were taken between the transmitting (#1) and each non-transmitting
channel (#2, #3 and #4) of a single actively SP4T switch, respectively. A schematic of the
experimental set-up is shown in figure 3.9, and the corresponding S21 measurements for
Channels #2, #3 and #4 are shown in figure 3.10, respectively. As indicated by the Sparameter measurements, each channel has greater than 40 dB of isolation across our
system’s operational bandwidth.
72
Figure 3.9. Schematic of the experimental set-up used to measure the isolation
between the active and non-active channels of a single active SP4T switch (shown
connected to channels #1 and #2). Isolation measurements for Channels #3 and #4
were measured using the same set-up (by replacing #2 with #3 and #4, respectively).
S21 measurements taken between the transmitting and non-active channels of a single
SP4T switch
Figure 3.10. S21 measurements of each non-active channel of a transmitting SP4T switch;
each channel has greater than 40 dB of isolation.
To demonstrate that the extra switch layer was responsible for the additional
isolation, S21 measurements were taken between the active channel of the first SP4T
switch in the network and the non-transmitting channels of an active switch in the second
73
level of our new bi-level network (schematic shown in figure 3.11). This measurement
set-up is representative of the previous system’s RF network configuration, and as shown
in figure 3.12, an additional 40 dB of isolation across the entire operational band was
obtained. However, the additional isolation seen in the 2nd generation system was
obtained at the cost of a significantly increased hardware footprint. Although originally
intended to employ all 32 channels in its design, the use of only 16 channels inadvertently
increased the system’s isolation characteristics.
Figure 3.11. Schematic of the experimental set-up used to measure the isolation
between the active channel of a 1st level SP4T switch and the non-active channels of
an active 2nd level SP4T switch (shown connected to Channels #1 and #3). Isolation
measurements for Channels #2 and #4 were measured using the same set-up (by
replacing #3 with #2 and #4, respectively).
A heterodyne receiver design was implemented to provide an excellent dynamic
range. The Agilent rfSG provided the desired transmissions signal along with a coherent
LO signal offset by 25 kHz and an additional coherent 25 kHz reference signal. Down
conversion of the transmitted signal with the LO provided a 25 kHz IF signal which
could be easily processed by the DAQ. Midrange IF signals allow the use of sharper
74
cutoff filters for improved selectivity, as well as the use of narrow band IF amplifiers
[102], limiting the need for additional frequency dependent componentry. We have
incorporated a custom narrow band IF amplifier in the current design to maximize
received signal power strengths as a result of the increased dynamic range of the 24-bit
signal acquisition boards.
S21 measurements
corresponding
the experimental
set-up
in2figure3.#.
Isolation Between
Trx Channelto
of SP4T
1 and each non-TRX
Channelshown
of a SP4T
0
Channel 2
Channel 3
Channel 4
-20
Amplitude [dB]
-40
-60
-80
-100
-120
0.5
1
1.5
2
Frequency [Hz]
2.5
3
9
x 10
Figure 3.12. S21 measurements taken between the transmitting channel of the 1st level
SP4T switch and the non-transmitting channels of the active 2nd level SP4T switch.
Although the received signals are highly attenuated due to traveling through the
breast and lossy coupling medium, any necessary amplification can be spread over the
system’s RF, IF and baseband stages to avoid potential signal instability, oscillation and
compression issues. Combined with the increased cost of high-frequency amplifiers, the
receiving channels’ component cascades were specifically designed to output a constant
75
low frequency signal to the DAQ in an effort to reduce back-end high-frequency signal
coupling that would result in the collection of corrupt data.
In half-duplex receiver designs, a channel’s transceiver unit does not act as a
transmitter and the receiver simultaneously. The duplexing can then be achieved by
connecting the channel’s antenna to the transmitting or the receiving path using a SPDT
switch. As mentioned earlier, we have incorporated an additional SPST switch in our
half-duplex design to increase isolation between the transmitting and receiving cascades
within the transceiver unit (figure 3.13.). The receiving pathway was designed to
intentionally demodulate the detected signal for the purpose of outputting a low
frequency IF signal to the DAQ. As a result, analyzing the effect of signal leakage from
the LO or RF networks on the IF signal’s output cannot be directly assessed using Sparameter measurements due to the limited dynamic range of the necessary measurement
equipment.
Figure 3.13. Schematic of the transceiver unit highlighting the incorporation of the
SPST switch used to increase the isolation between its transmitting and receiving
pathways.
76
In the previous design, the antenna-LO and RF-LO leakage paths were analyzed,
and high levels of network isolation were seen [41] (section 3.4). In advancing our
system, as with the development of any novel technology, the implemented design was
selected based on our experience regarding previous system development and its
resulting performance characteristics. Due to the lack of an existing isolation issue in the
previous RF switching matrix, the remodeling of its network inadvertently neglected
potential RF switch-based isolation issues. The purpose of streamlining the electronic
networks was to reduce and modularize the system’s design. However, this was achieved
at the unintentional cost of decreased isolation in the RF network.
Despite this design issue, the leakage is primarily associated with channels that
share a SP4T switch in the RF network. Its effect diminished as the distance between the
receiving and transmitting switches increases. Due to the sequential nature of the
system’s transmission strategy, whenever a channel is actively transmitting, the
remaining complementally channels can be referenced relative to the transmitter (figure
3.14.). For this leakage assessment, global channel #1 was transmitting and
measurements were made at relative receivers #1 to #15, respectively.
Figure 3.14. Schematic of the system’s circular antenna configuration highlighting the
transmitter-receiver distance variations (Compliments of Dr. Fox).
77
0
Leakage into Channel 2
500 MHz
900 MHz
1300 MHz
1700 MHz
2100 MHz
2500 MHz
2900 MHz
Received Signal Amplitude [dB]
Received Signal Amplitude [dB]
-20
-40
-60
-80
-100
-120
-140
0
-40
-60
-80
-100
-120
-140
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Relative Receiving Channel #
500 MHz
900 MHz
1300 MHz
1700 MHz
2100 MHz
2500 MHz
2900 MHz
Received Signal Amplitude [dB]
Received Signal Amplitude [dB]
-40
-60
-80
-100
-120
0
-40
-60
-80
-100
-140
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Relative Receiving Channel #
500 MHz
900 MHz
1300 MHz
1700 MHz
2100 MHz
2500 MHz
2900 MHz
-40
-60
-80
-100
-120
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Relative Receiving Channel #
(d)
Leakage into Channel 6
Received Signal Amplitude [dB]
Received Signal Amplitude [dB]
500 MHz
900 MHz
1300 MHz
1700 MHz
2100 MHz
2500 MHz
2900 MHz
-120
-20
-140
Leakage into Channel 5
-20
(c)
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Relative Receiving Channel #
(b)
Leakage into Channel 4
-20
-140
500 MHz
900 MHz
1300 MHz
1700 MHz
2100 MHz
2500 MHz
2900 MHz
-20
(a)
0
Leakage into Channel 3
0
Leakage into Channel 13
500 MHz
900 MHz
1300 MHz
1700 MHz
2100 MHz
2500 MHz
2900 MHz
-20
-40
-60
-80
-100
-120
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Relative Receiving Channel #
-140
(e)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Relative Receiving Channel #
(f)
Figure 3.15. Detected leakage signal through the RF switching matrix at channels (a)
two, (b) three, (c) four, (d) five, (e) six and (f) thirteen.
78
Figure 3.15 shows the measured leakage at the relative receiver numbers for
global receiving channels 2, 3, 4, 5, 6 and 13. Based on this investigation, it can be seen
that the leakage signal decreases as the distance from the transmitting switch source
increases (Channel 2 verses Channel 13). Furthermore, the circular design of the system’s
antenna array helps reduces the effect of this issue on the resulting signal measurements.
Antennas contiguous to a transmitting element receive signals with power levels similar
in magnitude to the transmitted signal’s power. This is due to their close proximity.
Fortunately, this is well above the leakage signal’s power level for frequencies up to 1500
MHz. Figure 3.16 plots DAQ measured signal amplitudes and the corresponding leakage
signals for every channel in the system over its entire operational bandwidth. The
corresponding SNRs are plotted and tabulated in figure 3.17 and table 3.1, respectively.
Overlay of the Measured Transmission Data and the System’s Noise Floor
Overlay of Measured Transmission Data and System Noise Floor Data
20
500 MHz meas
500 MHz noise
900 MHz meas
900 MHz noise
1300 MHz meas
1300 MHz noise
1700 MHz meas
1700 MHz noise
2100 MHz meas
2100 MHz noise
2500 MHz meas
2500 MHz noise
2900 MHz meas
2900 MHz noise
0
Received Signal Amplitude [dB]
-20
-40
-60
-80
-100
-120
-140
1
2
3
4
5
6
7
8
9
10
Relative Receiving Channel #
11
12
13
14
15
Figure 3.16. Overlay of DAQ measured transmission and leakage data for each
channel of the parallel detection scheme.
79
Signal to Noise ratios for each receiving channel
Signal to Noise Ratios for each Receiving Channel
500 MHz
700 MHz
900 MHz
1100 MHz
1300 MHz
1500 MH
1700 MHz
1900 MHz
2100 MHz
2300 MHz
2500 MHz
2700 MHz
2900 MHz
120
100
Received Signal Amplitude [dB]
80
60
40
20
0
-20
1
2
3
4
5
6
7
8
9
10
Relative Receiving Channel #
11
12
13
14
15
Figure 3.17. (a) SNR for each channel of the parallel detection scheme as a
function of frequency.
Table 3.1.
Table 3.1. Tabulated SNR [dB] for each channel of the system’s parallel detection
scheme
80
SNRs greater than 40 dB are seen up to 1500 MHz, and degrade as frequency and
the distance between the transmitting and receiving elements increases. This is expected
due to high frequency attenuation phenomenon and increased signal travel path lengths.
In the circular arrangement our antenna array, receiving channels 7, 8, and 9 are
positioned farthest from the transmitting element. Signals detected at these receivers
experience the largest amount of attenuation due to their increased travel-path lengths
from the transmitter through the breast and lossy coupling medium. Although these
channels are the limiting positions regarding optimal signal detection, SNRs greater than
20 dB are still achieved up to 2100 MHz. Furthermore, array elements located closer to
the transmitting antenna can achieve SNRs greater than 40 dB up to 2500 MHz.
By incorporating additional attenuating componentry into the RF network, our
system can achieve optimal inter-channel isolation, similar to the use of the additional
SPST switch in the transceiver units. Introducing an additional layer of SPST switches
with 40 dB of isolation into the RF network can limit the leakage through the switch
matrix configuration, at a limited addition to the overall electronics footprint. To confirm
the value of incorporating additional attenuating components into the switch network, a
second experiment was conducted to ensure leakage through the RF network could be
limited.
For this analysis, the RF connection to transceiver unit #2 was reconnected, and
40 dB attenuators were placed on the non-transmitting channels of the active switch. Data
was collected with the antenna port of transceiver #1 directly connected to the antenna
port of transceiver #13; all other antenna connections were terminated with matched
loads. Figures 3.18 and 3.19 show a schematic of the modified experimental set-up and
81
the corresponding measurements collected by the system, respectively. Introducing an
additional 40 dB of isolation into each non-transmitting channel of the active SP4T
switch eliminated the previously observed signals, increasing the system’s dynamic range
by ~20 dB and lowered the NF to approximately -110 dB.
Figure 3.18. Schematic of the modified experimental set-up incorporating an additional
40 dB of attenuation on each non-transmitting channel of the active switch (shown in
black).
82
0
Inter-channel Isolation
500 MHz
900 MHz
1300 MHz
1700 MHz
2100 MHz
2500 MHz
2900 MHz
Received Signal Amplitude [dB]
-20
-40
-60
-80
-100
-120
-140
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Relative Receiving Channel #
Figure 3.19. Plot of measured signals with channel #1 directly connected to channel
#13 (relative receiver #12), and all other channels terminated with 50 loads.
83
3.3 Measurement repeatability and effective operating bandwidth
To evaluate the quality of the system’s measurements, a breast-like phantom with
a tumor-like inclusion was repeatedly examined using the MI system. Data collected by
the DAQ was used to evaluate any resulting amplitude and phase measurement errors.
The phantom consisted of a glycerin:water mixture tailored to match the DPs of
heterogeneous breast tissue (86:14) and a high contrast tumor (50:50). The phantom was
submerged in the system illumination chamber, which was filled with an 80:20
glycerin:water coupling medium (figure 3.20). For frequencies up to 2 GHz, the average
error for each receiving channel was approximately 0.4% and less than 10 for the
measured amplitude and phase values, respectively. At higher operating frequencies,
system performance degraded, with the amplitude [dB] and phase errors reaching 0.8%
and greater than 20, respectively. The larger errors associated with high frequency
operation are presumably due to increased signal attenuation in the lossy coupling
medium as well as mixer performance issues associated with high frequency operation
when using stronger LO power signals [102].
Figure 3.20. Photograph of the breast phantom submerged in the imaging system’s
illumination chamber.
84
3.4 Additional isolation and leakage path analysis
We have already identified and addressed signal leakage through the modified RF
network. However, based on the error analysis, credible signal measurements can be
acquired up to 2 GHz. Additional leakage paths do exist between neighboring transceiver
units, and will be analyzed in this section.
As identified in [118], two possible signal leakage paths are present, each of
which can negatively impact inter-channel isolation (figure 3.21). These paths include
coupling between a transceiver unit’s (1) antenna and LO ports and (2) RF and LO ports.
As a result, received and transmitted signals can escape through these pathways and
ultimately leak into the IF output of a neighboring transceiver unit, due to their
connections in the PD network. Since both of these leakage route utilizes the PD network,
transceiver units on complementary 8-way PDs have increased inter-channel isolation
when compared to those that share the same PD. Additionally, the maximum leakage
signal strength occurs between neighboring transceiver units that utilize the same PD; its
strength degrades as the channel distance increases.
(1) Antenna-LO isolation: A -30 dB signal entering the antenna port of transceiver
unit #1 will be amplified by 36 dB as it travels through the receiving component cascade.
The mixer’s RF and LO ports have -20 dB of isolation, resulting in a leakage signal
strength of -14 dB prior to entering the LO amplifier. The reverse isolation of the LO
amplifier is -50 dB, allowing a theoretical leakage signal of -64 dB to enter the PD
network. S21 signal measurements taken between the LO and antenna ports of a single
transceiver unit are shown in figure 3.22. Overall, the measured data agrees with the
theoretical calculation. As this signals enters the LO PD network, it experiences further
85
attenuation. Unlike the previous system, the 8-way PDs have channel isolation greater
than 20 dB; the signal will exit the PD and be amplified by 18 dB at the neighboring
transceiver unit’s LO input amplifier. As the signal continues through the receiver
component cascade, it experiences an additional 40 dB of attenuation due to the mixer’s
LO-IF conversion loss of -40 dB. The resulting strength of the leakage signal into the
neighboring unit’s IF output will be less than -106 dB, an improvement when compared
to the previous system due to the increased channel isolation of the 8-way PD.
Figure 3.21. Schematic of two possible leakage paths through two adjacent transceiver
units: Antenna signals received by transceiver A leaked into the IF output of transceiver
B (green and red lines), and RF input to transceiver A leaked to IF output of transceiver
B (blue and red lines).
86
(2) RF-LO isolation: When a transceiver unit is in transmit mode, a 0 dBm signal
supplied to the RF port of transceiver unit #1 will be amplified by 18 dB as it enters the
transmission component cascade. In practice this power level is actually less than 0 dBm,
such that the amplified signal supplied to the antenna port is under 0 dBm due to patient
safety regulations. The isolation of the closed SPDT switch is roughly -50 dB,
consequently supplying the low noise amplifier a leakage signal of approximately -30
dBm. As it follows the receiving component cascade, the signal is amplified by 36 dB
and attenuated by 20 dB at the low noise amplifier and mixer’s RF-LO ports,
respectively. Continuing along the receiving pathway, the signal is attenuated an
additional 50 dB at the LO’s amplifier, resulting in a theoretical leakage signal of -64 dB
at the transceiver unit’s LO port.
S21 signal measurements taken between the LO and RF ports of a single
transceiver unit are shown in figure 3.22, and are consistent with the theoretical
calculation. As the signal enters the PD network it is subsequently attenuated by the 8way PD. Exiting the PD network, the signal is amplified by 18 dB and attenuated by 40
dB due to the neighboring transceiver unit’s LO amplifier and mixer LO-IF isolation,
respectively. The resulting strength of the leakage signal into the neighboring unit’s IF
output will be less than -108 dBm, which again, is an improvement when compared to the
previous system due to the increased channel isolation of the 8-way PD.
87
Figure 3.22. S21 measurements of a transceiver unit taken between its antenna and LO
ports (shown in blue), and RF and LO ports (shown in red).
88
3.5 Conclusion
We have identified two possible sources of signal leakage resulting from limited isolation
in the RF network’s configuration. Investigations into limiting the leakage signal were
successful, indicating that the incorporation of an additional level of SPST switches into
the RF matrix layout could alleviate this problem, at a limited increase to the overall
hardware footprint. Despite this signal, the analysis regarding the channel’s SNRs
including the leakage signal’s effect on the NF showed that SNRs greater than 40 dB can
be achieved up to 1500 MHz; SNRs for elements located closest to the transmitter can be
as high as 80 dB up to 2900 MHz. System sensitivity and error were assessed, with
results indicating reliable signal measurements and limited amplitude and phase errors up
to 2 GHz. Additionally, the integrity of the system’s measurements was investigated. The
results support the earlier finding that reliable magnitude and phase measurements can be
acquired up to roughly 2 GHz. Although the leakage signal is consistent, additional
electronic advances need to consider channel isolation issues in all electronic networks,
including possible RF switch leakage issues.
89
4. Motion Control System and Performance
The current system is superior to previous systems in its measurement capacity.
Increased motion control capabilities give the current system the ability to transmit (TX)
and receive (RX) both in-plane (IP) and cross-plane (XP) microwave signals, where
previous generation systems were limited to IP measurements only. This increased
measurement capacity was intended to allow 3D volumetric imaging, and combined with
our newly developed 3D reconstruction algorithm [99], Dartmouth’s Microwave Imaging
Spectroscopy Group is at the forefront of 3D clinical MT imaging.
Previous generation systems were restricted to IP measurement scenarios based
on their system’s design and selected motion-control technique (a single linear-actuator
controlled the entire antenna array). To increase the number of independent measurement
positions, the current system employs a strategy that independently controls the
movement of specific array elements to XP positions, while maintaining the ability to
reach traditional IP positions when desired. As a result of advancing to a bi-modal motion
control system, motion related performance characteristics have been assessed.
Signal S-parameter simulations will show that the amplitude of a detected
microwave signal significantly decreases as the vertical and horizontal distance between
the TX antenna and the complementary SA’s out-of-plane (OOP) RX elements increases.
We have already discussed the decreased power levels associated with array elements
that are located farthest from the TX antenna, and the resulting horizontal distancedependent affects to those channels’ SNRs (Chapter 3). This same phenomenon is
observed with vertical translation of the antenna arrays; as the linear-actuator controlled
mounting-plates approach their maximum separation during 3D data-acquisition,
90
transmitted microwave signals must travel through more of the system’s lossy coupling
medium on their path from the TX antenna to the OOP RX elements. This increased
distance is associated with additional signal attenuation. Furthermore, non-optimal beam
alignment of the TX antenna and the OOP RX elements at larger SA separation distances
directly affects the received signal’s strength.
The longer travel distance is associated with increased signal attenuation and
decreased signal strength at the OOP measurement sites such that for the larger
separation distances, these signals’ power levels reach and ultimately fall below the
system’s NF. This chapter will discuss the current system’s motion control technique,
which allows the acquisition of volumetric XP data for 3D image reconstruction. Signal
measurements taken with various SA spacing will be used to identify an optimal XP
separation distance, which will be frequency dependent. Phantom imaging experiments
have been conducted to visualize the effect of utilizing this long-range data in the
reconstruction algorithm. The results from this analysis will be used to guide the
development of a new position-optimized motion control system intended to allow faster
3D patient examinations by eliminating the need to acquire data the falls below the
system’s NF.
91
4.1 Independent sub-array control
The ability to independently control the system’s SAs is an important
advancement, allowing the collection of more patient data for improved image
reconstruction. The ability to obtain XP data is a significant clinical advancement;
utilizing this data in the reconstruction algorithm reduces model-data mismatch problems,
especially for the amplitude [99]. Additionally, when limited to 2D IP data acquisition,
small abnormalities that reside between two consecutive imaging-planes may not be
detected; the use of XP data acquisition can reduce this issue.
The increased measurement capacity is a result of an innovative illuminationchamber apparatus design and its subsequent motion control hardware arrangement
(figure 4.1a). Its ergonomic configuration enables independent SA movement within the
restricted space of system’s housing unit (figure 4.1b), allowing us to maintain system
modularity while increasing its overall measurement capabilities. The new design is low
(a)
(b)
Figure 4.1. Photographs of the imaging construct highlighting its (a) independent SA
mounting plates and antenna attachments, connected to the motion control system’s
linear actuators and (b) low–profile design positioned inside the patient housing unit.
92
profile when compared to it predecessor; 12.5 cm of vertical array translation can occur
while the patient is positioned on top of the system.
As discussed in Chapter 2, the motion system utilizes four linear-actuators driven
stepper motor complexes. The actuator–motor (AM) assemblies are symmetrically placed
underneath the illumination chamber, such that mirroring actuators are attached to the
same mounting plate (figure 4.1a). Consequently, the paired actuators are rigidly linked
to one another and the illumination array assembly. This physical attachment requires
coordinated, precisely timed and identical movement of the coupled actuators; unmatched
movements could potentially result in bending of the antenna array imaging construct due
to their large driving forces.
To avoid this situation, the motors are hardwired to each other through a four-axis
controller, with each assembly receiving power from a separate amplifier-controlling
unit. For each AM pair, an I/O output from an amplifier is connected to an I/O input of its
partner. Using this strategy, each motor within a pair will only execute movements if the
partner complex is functioning properly. The controller has a servo update rate of 62.5
axis. In the event that an AM complex stops operating properly, error signals are sent
to the partner’s controller and relayed to the corresponding amplifier drive, stopping all
motion upon signal reception.
In addition to the hardwiring of the two assemblies within a pair, they are coupled
through the motion control software using a similar strategy as the hardwiring technique.
The controller connects to the computer using an RS-232 connection, providing the
motion control program constant feedback signals from the controller at a baud rate of
19.2 kBd [119].
As a result, the computer is receiving 19,200 pulses
93
from the
controller, corresponding to pulse duration of 52
. Motion coordination is programmed
such that an individual AM assembly will only execute movements if its partner
assembly is not sending any error signals to the motion software, and due to the relatively
high communication rates between the computer, controller and motor, motion can be
stopped extremely fast. This strategy of linking the paired motors through both hardware
and software ensures identical motion execution while preventing an individual AM
assembly from moving when its partner is not.
The coordinated motion control of the paired assemblies allows independent SA
motion, permitting the acquisition of both IP and XC data. The definition of a particular
plane is based on a hardwired home position. These motors lack motion encoders;
consequently, optical sensors are used to define the home position at the beginning of
each motion sequence. The sensors are connected to the controller’s I/O channel, insuring
that each motor defines the same home position. All motion is made relative to the set
home position, which does not vary. A desired imaging plane can be reach using predetermined motion sequences where the user has the ability to select the number of
imaging planes and their spacing. These 2D and 3D motion strategies utilize an iterative
programming approach, both of which will be discussed in the following sections.
94
4.2 2D Movement strategy
2D data acquisition only utilizes IP measurement positions. After the motors have
defined their home position, the actuators raise the antenna SAs to the top of the imaging
tank. This is defined as plane 1 (P1), and all subsequent movements are made downwards
relative to the P1. In the 2D scenario, both SAs always reside in the same plane. The
procedure for a 7-plane 2D breast-imaging exam is as follows: both antenna SAs starts in
P1, and data is acquired as discussed in Chapter 2. Although the user has the ability to
select the number of desired transmitters, for clinical breast exams all array elements
transmit and receive, resulting in 240 independent measurements for each plane. After
data from P1 has been acquired, both SAs simultaneously move down to the second plane
(P2). The distance between the planes is selected by the user based on the size of the
patient’s breast while submerged in the illumination chamber. Once the SAs have
reached P2, data is acquired as before, resting in 240 independent measurements for P2.
The motion control program is iterative by design; subsequently, this procedure will
continue until both SAs have reached the final plane (in this case plane 7) and data has
been recorded for each transmitter. Once the imaging procedure has acquired data from
each plane, the motors re-home and the system waits for user input regarding execution
of the next motion sequence.
The program was designed iteratively due to the ease of utilizing FOR and
IF/THEN loops in LabVIEW. One drawback of this approach is that data from all planes
within the selected start and stop positions must and will be acquired. While this is not a
major issue in the 2D case, the iterative nature of the motion control program results in
non-optimized 3D data acquisition, which will be addressed in the next section. Figure
95
4.2a shows a photograph of a soda bottle phantom filled with an 86:14 glycerin:water
solution, including a 1 cm radius cylindrical inclusion made of a 50:50 glycerin:water
solution. These ratios have been tailored to roughly mimic the DPs of healthy and
malignant breast tissue, respectively. The phantom was placed in the imaging chamber
(figure 4.2b), and data was acquired for 4 in-plane positions using the method described
above. Figure 4.3 shows the 1300 MHz reconstructed permittivity and conductivity
images. The corresponding permittivity and conductivity values of the phantom were
measured using a dielectric probe (Agilent 85070E Probe Kit, Santa Clara, CA), and are
reported in table 4.1.
Table 4.1.
Region
Bath
Breast
Tumor
Center
[cm]
(0,0)
(0,0)
(0,0)
Radius
[cm]
11
5.5
1
Glycerin
%
80
86
50
Conductivity
Relative
Permittivity
[S/m]
22.1
1.22
14.1
0.98
53.089
1.43
Table 4.1. Measured dielectric properties of the phantom shown in figure 39.
(a)
(b)
Figure 4.2. Photograph of (a) phantom set-up and (b) phantom submerged in the
imaging system’s illumination chamber.
96
Plane 1
Plane 2
Plane 3
Plane 4
Figure 4.3. Four 2D IP reconstructed permittivity (top) and conductivity (bottom)
images (1300 MHz) of the phantom shown in figure 39, the phantom’s properties at
1300 MHz are given in table 4.1.
97
4.3 3D Movement strategy
The 3D data acquisition procedure uses the same iterative programming approach
as the 2D case. The user has the ability to select the desired operating parameters and can
initiate the data acquisition procedure. One addition to the 3D motion control program
allows the user to independently select the number of imaging planes for each SA,
although they are usually set to the same quantity to allow full coverage of the breast
volume.
The 3D analog to the 2D case previously described (7-planes of data) is as
follows: both SAs are raised to the top of the tank and P1 data is acquired as previously
described. The 3D program was designed to allow the acquisition of XP data in addition
to the traditional IP positions. After data has been acquired for P1, SA#1 moves to P2
while SA#2 stays in P1. Although P1 data was collected with both SAs IP, the movement
of SA#1 to P2 while SA#2 remains in P1 is representative of a XP acquisition task. Due
to the iterative programming approach, SA#2 will remain in P1 while SA#1 moves to
each position within the selected start-stop range; acquiring data at every location. Once
data for XP P1-P7 has been obtained, SA#1 will move back to P1 and SA#2 will move
down to P2. The iterative approach will proceed as before; SA#2 will remain in P2 while
SA#1 moves to each position within the selected range, with data being collected at every
site. After data for XP P2-P7 has been collected, SA#1 will move back up to P1 and
SA#2 will move down to P3. The procedure will continue until data has been acquired for
of every P3-Pt position, where t = 1:7 (based on the number of selected planes). Data
acquisition is terminated following the collection IP P7-P7 data.
Two issues arise as a result of using the iterative approach for 3D acquisition:
98
(1) The data acquisition process is quite long as a result of the SAs requirement to
acquire data at all possible Pt-Pt positions. For a 3D 7-plane examination, imaging times
are greater than 40 min/breast. One consequence of the long exam time is possible patient
movement, which can introduce artifacts into the reconstructed images. In addition, the
homogeneous data set used in the calibration process needs to be taken; this double’s the
operator’s time requirements. A goal of the work in this thesis is to increase 3D data
acquisition speed. This will allow shorter patient exams and lower operator time
requirements per patient, both of which are clinically attractive.
(2) The use of data collected at XP positions when the SAs are spaced farthest
apart can introduced artifacts into the reconstructed images because the measurements are
essentially noise. An additional goal of the work in this thesis has been to identify
optimal SA separation distances, consequently limiting reconstruction artifacts.
To illustrate this SA spacing phenomenon, a breast-like phantom filled with an
86:14 glycerin:water solution, including a submerged tumor-like inclusion made of a
15:85 gelatin:water mixture (figure 4.4), was reconstructed using varying amounts of
long-range data (2 XP, 4 XP and 9 XP); the reconstructed permittivity and conductivity
images are show in figure 4.5. Corresponding permittivity and conductivity values of the
phantom’s materials were measured using the dielectric probe, and are shown in table
4.2.
99
(a)
(b)
Figure 4.4. Experimental phantom set up showing (a) breast phantom with irregular
shaped inclusion placed in imaging tank (b) top-down view of phantom showing
breast and tumor phantom materials.
Region
Breast
Tumor
Conductivity
Relative
Permittivity
[S/m]
14.3
0.98
48.88
1.94
Table 4.2. Measured dielectric properties of the phantom shown in figure 4.4.
Table 4.2.
100
Figure 4.5. Reconstructed images (1500 MHz) of the phantom shown in figure 41 using
varying amounts of long-range data: (a) permittivity 2 XP, (b) conductivity 2 XP, (c)
permittivity 4 XP, (d) conductivity 4 XP, (e) permittivity 9 XP and (f) conductivity 9
XP.
This analysis was conducted to visualize the effect of long-range data on the
reconstructed microwave images. All figures utilized our 3D reconstruction algorithm
[99] and the same initial data. The only user-defined differences between the images are
the amount of XP data incorporated into the reconstruction process. The sets of images
were obtained using two consecutive XP (one above and one below) (figure 4.5a and
4.5b), four consecutive XP (two above and two below) (figure 4.5c and 4.5d) and a full
101
data (FD) set of nine XP incorporating all possible SA plane-combination (figure 4.5e
and 4.5f). From top to bottom, each set of images incorporates increasing amounts of
long-range data in the reconstruction algorithm.
As displayed in the recovered images, the use of XP information from the FD set
generated images with artificial features; it significantly overestimated the inclusion size,
position and property values in both the permittivity and conductivity images. However,
these undesired features were reduced as the amount of XP data decreased. By comparing
figures 4.5a and 4.5b to 4.5e and 4.5f, it is easily seen that the incorporation of data
acquired at large SA plane-separation distances resulted in degraded property recovery
and observable undesirable features.
As previously discussed, using the current motion system for 3D data acquisition
is accomplished by acquiring measurements at all possible SA plane combinations within
the selected range due to the iterative programming approach. Optimal reconstruction
data can be extracted from the non-optimized data acquisition session for post process
image reconstruction, as was done in this initial investigation. However, clinical
implementation of 3D imaging procedures has been limited due to the lengthy
examination times associated with the use of non-optimized motion sequences.
The advancement to clinical 3D imaging requires an analysis of SA planeseparation distances. Despite the ergonomically designed imaging construct and its highcaliber DAQ and microwave electronic programming, the current MT system has limited
flexibility regarding the measurement of transmitted data based on the selection of
optimal SA plane-separation distance. The larger plane-separation distances that occur
during current 3D data acquisition procedures are associated with increased levels of
102
signal attenuation. It is the author’s hypothesis that this signal attenuation phenomenon is
related to the misalignment of the TX antenna’s main beam and the complementary SA’s
OOP RX antennae at larger plane-separation distances. Additionally, elevated attenuation
levels are a result of the signals traveling through more of the system’s lossy coupling
medium as the plane-separation distance increases during current 3D data acquisition
procedures.
103
4.4 Sub-array plane-separation distance evaluation
During 3D volumetric data acquisition procedures, the distance between the
transmitting antenna and the OOP receiving antennae increases as the SAs reach their
maximum extension. Figure 4.6 displays the TX-RX plane-separation distances for the
farthest and closest IP and OOP antenna elements for complementary SAs, respectively.
The vertical displacements are intended to mimic the minimum and maximum possible
plane-separation distances for our imaging system.
(a)
(b)
(c)
(d)
Figure 4.6. TX-RX plane separation distances for (a) TX and farthest in-plane RX
antenna element, (b) TX and farthest out-of plane RX antenna element, (c) TX and
closest in-plane RX antenna element and (d) TX and farthest out-of plane RX antenna
element out-of-plane.
104
We have seen that the incorporation of microwave data obtained at larger TX-RX
plane-separation distances into the 3D reconstruction algorithm can result in the
appearance of artifacts in the reconstructed property images [99]. Furthermore, the use of
this long-range data is associated with deviations from the expected property values
based on phantom experiments of solutions with known dielectric properties [30]. To test
the hypothesis regarding beam alignment, HFSS (Canonburg, PA), amplitude simulations
were conducted using antennas modeled after of our system’s monopole design. Figure
4.7 show the electric field overlays for the antenna arrangement described in figure 4.6.
(a)
(b)
(c)
(d)
Figure 4.7. HFSS (Canonburg PA, USA) simulated electric field overlays of the TX
antenna elements for (a) TX and farthest RX in-plane element, (b) TX and farthest RX
out-of-plane antenna element, (c) TX and closest RX in-plane antenna element and (d)
TX and closest RX out-of-plane antenna element.
105
These simulation results indicate that signal measurements obtained at larger
plane-separation distances may be associated with the misalignment of the TX antennaelement and complementary SAs’ OOP RX antenna-elements. Although more
pronounced for the receiving elements that are located farther from the transmitter (figure
4.7a and 4.7b), the simulations indicate that at the larger plane-separation distances the
receiving elements may not be optimally aligned with the TX antenna element’s beam
pattern.
Image degradation resulting from the use of this long-range data can also be
associated with increased signal attenuation. The transmitted microwave signal must
travel through more of the system’s lossy coupling-medium as its detection path increases
during 3D data acquisition. Distance-related attenuation issues can decrease the
transmitted signal’s power level to such a point that is falls below the NF of the
transceiver unit’s receiving channel. The gain (G), noise figure (NF) and input noise (IN)
values for our system’s receiving channel are listed in table 4.3. As discussed in [36], our
system has an output NF of roughly -110 dBm.
Table 4.3.
System Parameter
Theoretically Calculated value
Input Noise [dBm]
-141 dBm
System Gain
28 dB
Noise Figure
5.35 dB
Table 4.3. Transceiver unit’s receiving channel’s electronic parameter values
106
(a)
(a)
(b)
(b)
Figure 4.8. Simulated S21 measurements for (a) TX and farthest RX antenna element for
various plane-separation distances (b) TX and closest RX antenna element for various
plane-separation distances.
107
In an additional analysis of signal attenuation related to larger signal travel paths
as the TX-RX plane separation distance increases, HFSS simulated S21 signal strengths
reach and ultimately fall below the NF of the system’s receiving channel. Reamplification of measured signals that fall below the NF of the system’s receiving
channel’s electronics decreases the accuracy of the signals measured by the DAQ. Figure
4.8 plots the simulated S21 spectrum of the complementary SA’s furthest OOP RX
antenna element (figure 4.7a) and the closest OOP RX antenna element (figure 4.7c) for
various TX-RX plane-separation distances, respectively. For the furthest RX antenna,
data obtained over the entire range of plane-separation distances (0 – 12 cm) seem to fall
below the NF of the system’s RX channel at roughly 1500 MHz, with larger separation
distances experiencing increased signal attenuation at lower frequencies.
The same trends appear in the S21 simulations representing the closest RX antenna
elements, although the phenomenon is less pronounced due to the shorter signal travel
distance. These simulations indicate that the current system is acquiring pre-amplified
signal measurements that are below the NF of the system’s receiving channels.
Consequently, a signal-sensitivity analysis in terms of frequency and SA spacing has
been conducted in an effort to define an optimal TX-RX plane-separation distance for use
in the newly developed motion-optimized 3D MT imaging system.
This iterative motion control strategy was, and continues to be quite effective for
2D imaging procedures where only in-plane data is required [25]. During 3D imaging
procedures, measurements are taken at all possible TX-RX plane combinations with the
intention of increasing the data available for image reconstruction. However, these
investigations have shown that the power levels of signals that traverse long TX-RX
108
travel paths do fall below the NF of the system’s receiving channel. These non-optimized
motion sequences unnecessarily increase clinical exam time by forcing the system to
acquire data that is easily corrupted due to previously mentioned signal attenuation
issues. Non-optimized data-acquisition procedures measure data at potentially
unnecessary SA plane combinations when using beam alignment and signal attenuation
criterion as metrics. Although optimal data can be extracted from the FD set, the iterative
nature of the current motion control program is associated with long data-acquisition
times.
To overcome these issues, hardware and software for a new motion system have
been developed and configured into a new MT prototype. This will allow the use of
signal-optimized motion sequences that acquire XP measurement data that is not affected
by beam alignment or additional attenuation effects. Optimizing motion sequences to
only acquire data when the TX-RX plane separation is such that all measured signals are
above the NF of the system’s receiving channel will increase exam speed with the
elimination of unnecessary measurement combinations. The elimination of these
measurement positions will allow 3D clinical data acquisition in feasible time frames.
109
4.5 3D IF signal and sub array plane separation distance analysis
To define optimal SA separation distances for receiving elements located on the
complementary SA relative to the transmitting element, an experiment was conducted to
determine the ratio of the measured IF signal’s strength to the noise floor as a function of
frequency and SA spacing. For this investigation, Channel 1 was chosen as the active
transmitting channel (figure 4.9-green). With the hypothesis that SA spacing influences
signal attenuation and the knowledge that receivers located farthest from the transmitter
experience the greatest amount of signal attenuation, specific high attenuation channels
positioned on the complementary SA relative to Channel 1 were chosen for this analysis.
They represented the most likely candidates to experience SA spacing dependent signal
attenuation, resulting in the detection of signals with power levels that fall below the NF
of the system’s receiving component cascades. All channels that reside on the same SA
as the transmitter were deemed unlikely candidates (figure 4.9-red), as their SA spacing
relative to the transmitter does not change during 3D data acquisition. As a result of
Channel 1 being designated the active transmitter, Channels 3, 5, 7, 9, 11, 13 and 15,
were disregarded due to being located on the same SA as Channel 1.
Based on the SNR analysis in Chapter 3, low attenuation channels with high SNR
were also deemed ineligible for this analysis, even if they reside on the complimentary
array. For example, although Channel 16 resides on the complementary SA, it has high
SNR characteristics (greater than 80 dB up to 2900 MHz), and consequently no useful
information regarding signal attenuation can be extracted from this channel. As
previously indicated, receivers located close to the transmitting channel detect signals
whose power levels are nominally on the same order of magnitude as the transmitted
110
signal’s power. This is supported by the SNR analysis performed in Chapter 3 as well as
the S21 simulations for the closest complimentary SA position (figure 4.8b), which in this
experimental set-up represents Channels 2 and 16, respectively (figure 4.9 yellow); due to
the symmetry of the circular antenna array. For similar reasons, Channel 4 was also
deemed ineligible based on its SNRs, which are greater than 30 dB up to 1900 MHz.
Figure 4.9. Schematic of experimental set-up showing the transmitting element
(green), ineligible IP elements (red), the near neighbor (NN), middle neighbor (MN)
and far neighbor (FN) receiving channels (blue) and their symmetric counter parts
(white).
Specific near neighboring, middle neighboring and far neighboring channels were
selected for this investigation, each of which has high attenuation characteristics as a
result of their location relative to the transmitter and their low SNRs as indicated in
111
Chapter 3’s analysis. As previously noted, the receivers located farthest from the
transmitter represent the system’s limiting channels in terms of measurement integrity,
especially at low input power levels and high operational frequencies. This is confirmed
by the simulated S21 measurements of the furthest spaced OOP receiver (figure 4.8a),
which in this experimental set-up corresponds to Channel 10. Due to the fact that
Channel 10 (R10) represents the maximally separated receiving channel and is the
physical equivalent of the S21 measurements shown in figure 45a, it was selected as the
far neighbor channel for this investigation. Channel 12 (R12) was selected as the middle
neighbor due to its degraded SNRs and its location relative to the far neighbor and the
transmitter. R10 and R12 both reside on the complementary SA; as a result, SA spacingdependent signal attenuation characteristics can be extracted by analyzing these channels.
Receiving Channel 14 (R14) was selected as the NN; it was the closest remaining eligible
channel that resides on the complementary SA. Based on the symmetry of the array,
Channels 6 and 8 are the mirror equivalents of Channel 10 and 12, respectively. As a
result, their sensitivity can be inferred by assessing R10 and R12. Having selected the
channels of interest (figure 4.9 blue) based on their location relative to the transmitter and
their SNR characteristics, the IF outputs from these channels were independently
assessed to evaluated their SA spacing-dependent signal attenuation.
We can utilize the S21 simulation results for the farthest receiving channel (figure
4.8a) to develop optimal SA spacing metrics. Its data corresponds to R10 in this
investigation. Since R10 is the limiting factor in terms of SA separation-related
attenuation due to its placement relative to the transmitter, which was supported by the Sparameter simulations and horizontally confirmed using the SNR analysis in Chapter 3,
112
determining optimal SA separation distances for each channel can be based on R10’s
sensitivity to SA separation distance. By utilizing the maximum R10 SA spacing where
all received power levels are above the NF as a guide, an operational metric has been
development as a function of frequency for each SA spacing distance.
The optimization of spatially-dependent signals are limited by the channel that
experiences the largest spatially-dependent attenuation; subsequently Channel 10 served
as the baseline for developing this metric. We know from figure 4.8a that data collected
with a SA separation distances greater than 6 cm will fall below the NF of the receiving
component’s electronic cascade at frequencies above 1500 MHz. For any frequency
above 1500 MHz, R10 will be receiving signals that fall below the receiving component’s
NF. Since the goal of this investigation was to define optimal SA spacing such that all
received signals for every channel are above the NF and knowing R10 is the limiting
channel, its uppermost frequency where uncorrupt data can be collected represents the
limiting operational frequency. Consequently, 1500 MHz was used as the uppermost
optimal frequency in developing this metric for the near neighbor, middle neighbor and
far neighbor. Analyzing the ratios of the desired IF signal to the noise floor of the near
neighbor, middle neighbor and far neighbor channels at 1500 MHz provided the limiting
criteria for defining optimal SA spacing.
Using the DAQ, output IF signal were measured at these channels as a function of
frequency (1100 MHz to 2300 MHz in 200 MHz increments) for 11 XP positions, each
separated by 1 cm. By subtracting the measured noise from the desired signal, SA
sensitivity metric tables have been developed for the far neighbor, middle neighbor and
near neighbor channels, and are shown in tables 4.4 – 4.6, respectively.
113
Table 4.4.
Table 4.4. Tabulated SA sensitivity results for the far neighbor (Channel 10)
Table 4.5.
Table 4.5. Tabulated SA sensitivity results for the middle neighbor (Channel 12)
114
Table 4.6.
Table 4.6. Tabulated SA sensitivity results for the near neighbor (Channel 14)
Results from this analysis are very informative. Each of the investigated channels
show a similar trend; as the XP separation distance increases, the attenuation of the
detected signals increases, as indicated by the decreasing sensitivity values associated
with the larger XP separation distances. Furthermore, this analysis shows that the
attenuation is frequency dependent; as the frequency increases, the sensitivity ratios
decrease.
Additionally, it is observed that channels located furthest from the transmitting
antenna experience additional signal attenuation. By comparing the sensitivity values of
the far neighbor to the near neighbor, it is easily seen that signals detected by RX
elements that are located further from the transmitter experience increased attenuation.
An interesting result of this study is seen for the lower frequency measurements of the
middle neighbor; when compared to the far neighbor and near neighbor, the middle
neighbor’s values are rather consistent for the lower frequency ranges. Although one
115
might assume this is related to the decreases attenuation associated with lower frequency
operation, the same trend in not seen in the near neighbor or far neighbor results, both of
which experience broad decreases at these frequencies. Consequently, the limited
attenuation associated with the middle neighbor at low frequencies may be due to
enhanced beam alignment when compared to the near neighbor and far neighbor.
Even though the near neighbor’s low frequency sensitivity values are higher than
both the far neighbor and middle neighbor, its range of values is larger than both the far
neighbor and middle neighbor, respectively. This indicates that vertical translation of the
near neighbor antenna is significantly impacted by beam alignment as opposed to signal
attenuation due to increasing travel path. At XP 11, the TX and RX are separated by 12
cm, which is roughly the same separation distance as the transmitter and the far neighbor
for XPs 1 and 2, respectively. However, for the far neighbor at those positions, the low
frequency attenuation ranges are not as broad when compared to the near neighbor
channel. Although distance-related attenuation is observed in this analysis, beam
alignment is an issue, especially for closer antenna elements. At shorter XP distances, the
near neighbor is in proper alignment with the transmitter’s beam, and as the near
neighbor’s vertical XP distance increases, the beam is no longer optimally aligned.
While this is an interesting finding, this phenomenon will not impact the resulting
data at this measurement site because its values are well above the receiving channel’s
NF. However, for the middle neighbor and far neighbor, the decreased sensitivity values
are more heavily impacted by frequency, although XP distance does add to their
diminished signal detection strength. For six cm of data (which represents seven imaging
planes with a 1 cm slice thickness), the simulated S21 measurements suggest that non-
116
corrupt data can be obtained at R10 for any frequency less than 1500 MHz. Using the
sensitivity metric tables for the near neighbor, middle neighbor and far neighbor in
association with the cut-off frequency of 1500 MHz, we can define “out of bound” XP
positions that will result in the collection of corrupt data. Based on 6 cm of SA spacing
we can see:
(1)
Any near neighbor with a sensitivity ratio < 82 will be “out of bounds” and as
a result, diminished signal strengths will be detected at the middle neighbor
and far neighbor.
(2)
Any middle neighbor with a sensitivity ratio < 40 will be “out of bounds”
resulting in the detection of diminished signal strengths at the far neighbor.
(3)
Any far neighbor with a sensitivity ratio < 31 will be “out of bounds” resulting
in the collection of corrupt data at that measurement site.
These metric tables can be used to define optimal SA separation distances
provided that none of the above “Out of Bound” values are used, theoretically allowing
the detection of reliable data. For example, if 7 XP of data are desired, the maximum
frequency that will allow the collection of reliable data is 1300 MHz. At 1500 MHz, the
far neighbor’s sensitivity value of 28 indicates an “Out of bound” measurement position
for this frequency. Looking back at the 3D reconstructed images in figure 4.5, we can
now understand why the images degraded as we incorporated more XP data. Those
images were constructed at 1500 MHz; using the far neighbor metric for 1500 MHz, the
maximum “In bound” number of XPs is six. When the FD set including 9 XP was
117
reconstructed at 1500 MHz, we saw the generation of artificial features in the recovered
images. Inspecting the far neighbor metric table we can deduce two key points:
(1) Reconstructing 9 XP at 1500 MHz represented an “Out of Bound” position
(23 < 31)
(2) To reconstruct that FD set with reliable data based on the far neighbor metric,
the optimal reconstruction frequency is 1300 MHz (32 >31)
Figure 4.10 shows the 1500 MHz FD permittivity image and the corresponding
1300 MHz FD permittivity image. The reduction of artifacts is immediately apparent, and
is presumably due to reconstructing the data using optimal data, which is distance and
frequency dependent. Results from this investigation indicate that optimal XP distances
can be identified using the SA sensitivity metrics created for this analysis.
(a)
(b)
Figure 4.10. Reconstructed permittivity images of the phantom shown in figure 41 (a)
at 1500 MHz using FD and (b) at 1300 MHz using FD.
118
4.6 Conclusion
The current lack of freedom in selecting optimized motion sequences based on
optimal TX-RX plane positions and the large time required for acquiring non-optimized
3D data sets (> 40 min / breast for a traditional 7 plane imaging sessions), combined with
the current availability of low-profile motion encoders have been driving forces in the
development of a new MT prototype that incorporates an updated motion-control system,
allowing user-friendly, clinically-feasible and patient safe 3D microwave-imaging
procedures. In an effort to minimize exam time and produce optimized motion sequences
by eliminating the need to measure data that is below the NF of the system’s receiving
channel, a TX-RX plane-separation sensitivity analysis has be conducted to determine the
optimal transmit-receive plane-separation distance. To increase exam speed, signaloptimized motion-control algorithms have been developed to control the new motion
system, permitting motion-optimized data-acquisition based on the optimal SA separation
distances.
119
5. Motion Control System Improvements
To overcome the signal attenuation issues related to the increasing SA planeseparation distance that occurs during non-optimized 3D data acquisition, a new motion
system has been created that is capable of executing optimized motion sequences. The
increased functionality of the new motion system allows the collection of optimal
reconstruction data during the data acquisition process as opposed to extracting it from a
large non-optimized data set during post process image reconstruction. The use of
optimal SA plane-separation distances, in conjunction with the ability to select optimized
data-acquisition sequences as a result of the newly developed motion system has
decreased clinical 3D examination times while providing the necessary data for accurate
image reconstruction.
(a)
(b)
Figure 5.1. New motion system highlighting (a) placement of rotary motor-controlled
linear-actuators and motion encoder placed in data acquisition structure and (b) flash
programmable amplifier drives and AC/DC power converter.
120
The new motion system incorporates low profile motion encoders and flash
programmable amplifier drives which have been assembled and retrofitted to reside
within the ergonomic construct of a new illumination chamber-imaging construct (figure
5.1) modeled after the apparatus described in Chapter 4. A series of communication
techniques were evaluated in an effort to create a highly functional and user-friendly
motion system.
121
5.1 Updated Motion Control System
The key hardware additions of this new design are flash programmable amplifiercontrol drives (Copley Stepnet Digital Drive STX-115-07) and the placement of lowprofile motion encoders (Parker Custom Encoder Products 260 series) on the bottom of
each AM assembly. Each motion encoder and amplifier controller combination has been
arranged to allow two-way communication between the computer, amplifier and the
encoder; by triggering a motion command from the computer, particular positions can
easily be reached as a result of the encoder’s ability to track the exact extension of the
actuator’s arm. Controllers receive feedback from its associated encoder regarding the
actuator’s current location; permitting SAs position information to be relayed from the
encoder to the amplifier controllers and the motion control software. This permits direct
knowledge of the SAs position inside the illumination chamber, allowing each SA to
individually move to any vertical location within the range of the actuator’s extension
length (12.5 cm).
Each amplifier-controller has flash programmable memory capabilities, allowing
the storage of predefined motion sequenced that can be triggered through the controlling
computer. With this new design the home position of the AM pairs can be stored on the
amplifier drives, and upon executing a motion sequence home position can be reached as
a result of the motion encoders’ ability to track the actuator arm’s current extension.
Although the on-board memory allows the storage of the system’s home position as well
as a minimum and maximum extension length, optical sensors can still be used to define
home and set upper or lower actuator arm extension distances; consequently no AM
complex will exceed its programmed extension range.
122
We have re-implemented the dual independent control of the SAs in this design;
however, with the use of the encoders the new motion control software has the freedom
to move each AM pair to any vertical array position within the illumination chamber,
non-iteratively. Care has been taken in defining the home, maximum and minimum
position locations; consequently, all movements are now made globally due to the
system’s knowledge of the home position. The spacing between the SA mounting plates
and the different lengths of the non-active parts for each SA element have been carefully
chosen to allow unimpeded movement of each SA such that SA#1 can traverse all 12.5
cm of its possible vertical translation length while SA#2 is located at any plane within the
illumination chamber or traveling its entire vertical translation length, respectively. This
motion innovation is critical; now the system has the ability to acquire data at any userdefined position within the illumination chamber as a result of developing a non-iterative
motion control algorithm.
The new amplifiers have data transmission rates of 115 kBd, approximately one
order of magnitude greater than the previous system (~19 kBd). Communication with the
controlling PC can be accomplished utilizing a RS-232 or controller-area network (CAN)
connection, and can interface with LabView. Each amplifier drive receives digital
encoder feedback at 20,000,000 counts/s and has twelve digital I/O inputs; allowing the
hardwiring of the paired AM complexes as described in Chapter 4. Linking the drives
through the motion control software has also been reemployed in the new software.
Additionally, the drives are equipped with bicolor LED displays that indicate operational
status by color and blinking [120]. Connection between the amplifier drives is
accomplished using Ethernet cables, allowing information to be relayed from the
123
computer to the controllers and vice versa. Using these amplifiers in conjunction with the
stepper motors, a maximum of 100,000,000 steps per revolution can be achieved.
124
5.2 RS232 communication controlled with CME2 software
Initially, motion sequences were saved to the amplifier drives’ flash memory
using the CME2 software package (Copley Controls, Canton, MA 2009). This software
contained a user friendly GUI that allows the development of motion sequences based on
available internal functions that allow homing, relative and global movement of the
actuators. Using an RS-232 connection, the amplifiers can be triggered to execute the
preprogrammed motion sequences using the CME2 package. Figure 5.2 displays a
screenshot of the CME2 GUI with more than 20 preprogrammed motion sequences
available for selection.
Figure 5.2. CME2 software GUI utilized for developing motion sequences that are
stored in the amplifier’s flash memory drive. This image shows more than 24 different
motion sequences, each of which can be executed through the CME2 software or an
ASCII command exporter.
125
While this motion control technique does allow independent SA movement to any
possible location within the defined maximum and minimum actuator range, using the
CME2 software to trigger motion execution occurs outside of the system’s main control
program, which is undesirable for our integrated system. To incorporate triggering of the
stored motion sequences into the system’s main control program, a custom ASCII export
program was created in LabVIEW.
126
5.3 RS232 communication control with LabVIEW software
Utilizing the serial ASCII export program was a successful solution for triggering
the custom motion sequences through the main control program. Although this technique
can allow the storage of all the possible motion sequences we may desire, it lacks the
freedom to change any desired parameter through the user interface without altering an
existing motion sequence and re-storing it in the amplifier’s on-board memory. This is
undesirable, especially in clinical settings where motion sequence parameters such a slice
thickness and the number of desired imaging planes are patient specific and may require
immediate changes. Additionally, RS-232 communication addresses each amplifier in
series; as a result, preprogrammed time delays were needed to ensure that the coordinated
motion of all linear actuators occurred simultaneously.
The use of preprogrammed time delays is unsuitable for our design; the smallest
timing offset can result in contortion of the ridged construct as the actuators are
physically locked together via the mounting plate arrangement. Based on our desire to
allow the selection of user-defined motion sequences with the ability to change specific
motion parameters in real-time through the system’s main control program, a LabVIEW
VI that controls every aspect of the AM complexes including all motion system
procedures and parameter selection has been developed. Additionally, to eliminate the
RS-232 associated time delays, a CAN adapter (Kvasier Leaf Lite) has been employed
for communication between the PC and amplifier drives (figure 5.3).
127
Figure 5.3. Kvasier Leaf Lite HS Controller-Area Network (CAN) to Universal Serial
Bus (USB) adapter (Mission Viego, CA).
128
5.4 CAN adapter communication controlled with LabVIEW software
The CAN adapter is capable of addressing all of the amplifiers in a parallel
fashion. Its incorporation, combined with the use of the new LabVIEW motion control VI
has eliminated the need for both preprogrammed motion sequences and time delays
associated with the use of RS-232 communication and preprogrammed motion control
triggering, respectively. The ability to address all amplifier drives simultaneously has
allowed the development of a motion program capable of independently moving both
SAs to all possible position combinations within the actuator’s programmed minimum
and maximum extension range. The freedom to select any desired TX-RX plane
combination allows the execution of optimized data acquisition sequences based on ideal
TX-RX plane-separation distances.
Figure 5.4 shows the GUI of the newly developed motion control program, which
is capable of selecting any TX-RX plane combination within the selected set-up
parameters (shown for 7 slice planes with a 1 cm slice thickness). The motion control
matrix allows the selection of any TX-RX plane combinations by clicking the
corresponding box in the motion control position matrix (highlighted in figure 5.4). For
example, a traditional 2D seven-plane imaging sequence would have the diagonal of the
7x7 motion matrix filled, with all other boxes empty. Optimized data acquisition
sequences based on ideal TX-RX plane-separation spacing has increased clinical exam
speed; specifically eliminating the time associated with measuring signals that fall below
the NF of the system’s receiving channel, which as discussed in Chapter 4 is a result of
the increased plane-separation distance that occurs during non-optimized 3D data
acquisition.
129
Sub Array 1
Position Matrix
Sub Array 2
Figure 5.4. New motion system’s LABVIEW user interface showing a 7x7 motion
combination matrix.
130
5.5 Conclusion
A new MT imaging system has been constructed; it contains an updated motion
system that is capable of selecting optimal data acquisition sequences. Experiments
conducted in Chapter 4 evaluated SA plane-separation spacing; the results have been a
primary factor in designing the new motion control system. This optimization has
eliminated the acquisition of XP positions that are known to result in the collection of
data that falls below the receiving channel’s NF. The SA separation analysis and the
resulting construction of the new motion control system allows customized and noniterative optimized data acquisition procedures; increasing patient examination speeds by
eliminating the requirement to collect data at XP combinations where very low powered
signals are detected. This allows optimal reconstruction data to be collected during the
patient examination as opposed to extracting optimal data from a non-optimized FD set.
As noted before, this is a significant clinical advantage for both the patient and the system
operator.
131
6. Integrating updated motion control system and mounted optical
scanner
To overcome the signal attenuation issues related to the increasing SA planeseparation spacing that occurs during non-optimized 3D data acquisition, a new motion
system has been created that is capable of executing optimized data acquisition
sequences. The increased functionality of the new motion system allows the collection of
ideal reconstruction data during the data acquisition process. This allows the collection of
data for enhanced image reconstructing without the need to select optimal microwave
data from large non-optimized data set during post process image reconstruction. The use
of optimal SA plane-separation spacing, in conjunction with the ability to select
optimized data-acquisition sequences due to the newly developed motion system has
decrease 3D clinical data acquisition times while providing the necessary data for
accurate image reconstruction.
In addition to the use of optimal cross-plane SA separation spacing, we have seen
increased accuracy in the recovered dielectric property distributions of reconstructed
microwave images resulting from the incorporation of the OUT’s boundary information
into the reconstruction algorithm. This information is used to generate a custom
boundary-derived mesh for use in a conformal microwave imaging reconstruction process
(Chapter 7). Boundary measurement information is easily obtained in simulation and
phantom experiments for regular geometric shapes. For more complex shapes, alternative
means of accurately measuring the OUT boundary are necessary.
We have recently shown that boundary information from MR images can be used
in an MR-guided CMI reconstruction process [121]. The irregular boundary can be
132
extracted from the MR image, and a boundary derived reconstruction mesh can be
generated using custom mesh generation software written in MATLAB. MR images
reflecting the buoyant nature of the breast while submerged in the MT system’s dense
coupling medium are needed for the extraction of accurate breast boundary information.
Consequently, we have developed a prototype MT imaging system that can operate
within the high-strength magnetic field of the MR system’s magnetic bore [122] (figure
6.1).
(a)
(b)
Figure 6.1. MR compatible MT imaging chamber’s (a) antenna array connected to semiridged coaxial cable and (b) placed on the MRI system’s patient positioning table.
The use of this combined system allows accurate extraction of irregularly shaped
boundaries from the high-resolution MR images while simultaneously obtaining the
corresponding MT data. Attachment of the MR compatible illumination chamber is
accomplished by removing the system’s non-optimized 3D imaging apparatus and
attaching the MR array to the system’s bulkhead mounted SMA connectors (figure 6.2).
133
This eliminates any unnecessary patient movement and/or repositioning, which may
decrease registration accuracy between the two imaging modalities if the imaging
procedures were performed separately. Initial clinical results from the MR guided MT
system have recently been reported [101].
(a)
(b)
(c)
(d)
Figure 6.2. Photographs of (a) MT system’s electronics attached to wall mounted
bulkhead panel separating shielded MRI room from control room, (b) MT system’s
mounted SMA adapters, (c) wall mounted bulkhead panel inside MRI suit and (d) MR
compatible MT imaging array positioned inside the bore of the MRI system.
The combined MT-MR system is in preliminary stages of development; currently,
only a single 2D tomographic slice of in-plane data can be acquired. As a result,
alternative means of extracting the needed boundary information during clinical 3D MT
134
data acquisition procedures have been explored. MIS group member Matt Pallone has
developed a mounted optical scanning system for the purpose of measuring the boundary
of objects while submerged in the 3D system’s illumination chamber [123]. Output from
the scanner is used to create a custom reconstruction mesh for objects with both
geometrically and irregularly shaped boundaries; its use allows the generation of a
custom reconstruction mesh for each OUT imaged in the 3D system.
135
6.1 Master and Slave Computer System Communication
A new imaging apparatus incorporating the updated motion control and optical
scanning systems has been built. The hardware for both of these new systems has been
assembled and fit into a prototype-imaging construct (figure 6.3) modeled after the
original imaging system [30]. The main difference between the old and new imaging
constructs are the introduction of motion encoders and flash programmable amplifier
drives to the motion system hardware, and the rigid attachment of the optical scanner to
the imaging apparatus. As previously stated, the system is highly modular, facilitating the
replacement of the system’s illumination chamber with a variety of imaging arrays.
Connection of the new optimized imaging apparatus to the current system’s microwave
electronics and DAQ systems has been achieved by employing the same technique
utilized for connecting the MR compatible imaging chamber.
Figure 6.3. Photograph of the newly developed MT imaging prototype showing
mounted optical scanner.
The new motion and optical scanning systems’ hardware is controlled by a
separate computer than the microwave electronic and DAQ systems. Communication
136
between these two computer systems has been achieved using an RS-232 connection in
conjunction with LabVIEW’s Virtual Instrument Software Architecture (VISA) controls.
The establishment of this closed network, based on the serial connection and VISA
controls, requires the assignment of a Master and Slave computer in the communication
initialization process. We have designated the motion/optical scanning system’s computer
as the Master computer based on its increased computational power and available disk
space. The existing MT computer has been designated as the Slave computer, executing
its programmed operations only when triggered by the Master system.
137
6.2 Master Computer Control and Operation
The Master computer controls the newly-developed antenna motion and optical
scanning systems as well as established operational functions such as the pumping of the
coupling-medium into the illumination tank. Communication between the Master and
Slave computers is initiated through an active handshaking process. Once communication
has been established, trigger signals are sent from the Master to the Slave computer
system instructing the Slave system to execute its programmed responsibilities.
The electronic hardware associated with the Master system has been incorporated
into a new patient bed (figure 6.4a) placed directly next to the current electronic and
DAQ systems (figure 6.4b). This positioning reduces the distance between the new
construct’s antenna array elements and their corresponding transceiver units in the
measurement system, minimizing signal attenuation issues associated with the use of
(a)
(b)
Figure 6.4. (a) Hardware for the updated motion control system and optical scanning
system placed in electronics cart underneath the new patient bed and (b) new
illumination apparatus and associated hardware placed next to current microwave
electronics’ system.
138
longer cables. Figure 6.5 displays the Master computer’s control GUI. Once
communication has been established, examination parameters such as frequency range
and interval, slice thickness, array positions and the desired number of TX antenna
elements are entered into the Master computer’s control interface. The system is set-up to
automatically pass the appropriate examination parameters to the Slave control program.
Figure 6.5. Master computer system’s GUI ready for the user to enter examination
parameters.
139
6.3 Slave Computer Control and Operation
After the Slave control program has been started, communication with the Master
computer is initiated by pressing the GUIs uplink button (figure 6.6). During the
communication initiation process, the Slave computer waits until a signal is received
from the Master computer indicating system communication is operational. Once
communication has been established, the Slave computer waits for input parameters to be
sent from the Master control program. This information is used in the operation of the
Slave system’s microwave electronics and data acquisition programs.
Figure 6.6. Slave computer system’s GUI prior to the establishment of
communication with the Master system.
140
When triggered, the Slave computer is responsible for executing microwave
signal transmission, signal modulation, measurement extraction and file storage
procedures. The measurement data is then passed back to the Master computer for later
use in our off line reconstruction procedure. The Master-Slave arrangement has been
advantageous; allowing us to maintain the use of the current high-integrity microwave
data acquisition system while incorporating the new motion optimized scanner-mounted
imaging apparatus.
LabVIEW’s VISA controls facilitate communication between the two computer
systems. This technique permits two-way communication between the Master and Slave
computer systems allowing bi-directional signal triggering and data transfer. Figure 6.7
shows the Slave GUI after communication between the two computer systems has been
established. Upon initiation of the data acquisition procedure, the Master control program
moves the antenna SAs to their starting positions. Next, a trigger is sent to the Slave
computer initiating the execution of microwave signal transmission, modulation and
measurement and storage protocols. This initial round of data acquisition is followed by
the transmission of a signal from the Slave computer back to the Master indicating data
acquisition for the current array position has concluded. If the selected data acquisition
procedure’s motion sequence requires more than one imaging position, the Master
program then moves the antenna SAs to their next position. This is followed by the
transmission of a signal from the Master computer back to the Slave triggering the next
round of data acquisition. The back and forth communication between the two systems
continues until signal measurements have been recorded for all of the selected SA
positions.
141
Figure 6.7. Slave computer system’s GUI after the establishment of communication
with the Master system.
142
6.4 Conclusion
An optimized motion system has been constructed and mated with a newly
designed mounted scanning system. This combined system has the ability to acquire IP
and XP data with optimal SA spacing in addition to extracting the boundary of the OUT
while submerged in the system’s illumination chamber. The system has been placed next
to the current data acquisition system and connects to its transceiver units through the
mounted SMA connectors utilized for attaching the previous non-optimized imaging
apparatus. Communication between the two controlling computer systems is
accomplished using an RS-232 connection, with the new and old computer systems
taking on the Master and Slave roles, respectively. This new arrangement allows
optimized data acquisition procedures in significantly shorter time frames in addition to
extracting the boundary of the OUT during the examination procedure. The critical
advances for this work include: increasing 3D data acquisition speed, allowing 3D patient
imaging in clinical feasible time frames while enhancing the resulting reconstructed
images by utilizing patient specific reconstruction meshes generated by the optical
scanner (Chapter 7).
143
7. Conformal Microwave Imaging
We have previously shown that conformal microwave imaging (CMI) has the
ability to improve the accuracy of recovered dielectric property distributions when a
saline-based liquid is used as the coupling medium [124]. In this two-part reconstruction
procedure the breast boundary was extracted from a microwave image and used to create
a boundary derived reconstruction mesh for the CMI reconstruction process. This
technique relied on the relatively large contrast that existed between the coupling medium
and OUT (i.e., breast) in order to extract the breast boundary from the microwave image.
More recent experiments have indicated our log-transform reconstruction
algorithm increases the accuracy of recovered dielectric properties and has improved
convergence characteristics over previous techniques [94]. Image accuracy has also
benefited from the use of a dielectrically tailored glycerin coupling-medium that
suppresses unwanted multipath signals [122]. This tailored coupling-medium reduces
reflections at the OUT-bath boundary in an effort to increase scattering at dielectrically
contrasting regions inside the OUT.
It has recently been shown that dielectric property recovery can benefit from the
incorporation of OUT boundary information of objects with geometric [125] and
irregular [121] shaped boundaries when using a lossy glycerin/water-coupling medium.
However, using the dielectrically tailored glycerin coupling-medium obscures the OUTbath boundary in the reconstructed image, rendering the two-part microwave-driven CMI
reconstruction process unfeasible (i.e. we essentially use the contrast to visually assess
where the boundary is). We have overcome this issue with the use of the new imaging
system’s optical surface scanner [123]. The scanner is used to extract the desired
144
boundary information required for the CMI reconstruction process.
This chapter will focus on the conformal meshing technique in microwave
imaging, presenting both simulation and phantom results that demonstrate the power and
effectiveness of using the approach with our current imaging system; for both 2D and 3D
image reconstruction. Results from this investigation will show a correlation between the
reconstructed image’s quality and the size of the utilized parameter mesh; conforming
this mesh to the approximate size of the OUT’s perimeter results in optimal property
recovery.
145
7.1. 2D Simulation Studies
A series of simulation investigations have been conducted to evaluate the efficacy
of utilizing the CMI reconstruction process with our current system and 2D
reconstruction algorithm. Three 2D simulation cases (each including -100 dBm of
additional noise) were evaluated at 1300 MHz, each of which compared the uniform
reconstruction (7 cm radius property mesh) to its corresponding conformal counterpart
(5.5 cm radius property mesh) for both the permittivity and conductivity images. The
three cases consisted of a homogeneous background (
with a breast-like region (
and
18 and
22.3 and
1.22 [S/m]),
1 [S/m]) and tumor-like inclusion (
50
2.0 [S/m]). Case #1 contained a single inclusion region, where cases #2 and #3
contained two and three inclusion regions, respectively.
Each case utilized a FE
algorithm described in [117] to compute the forward solution and the reconstructed
images were obtained using our 2D FDTD algorithm [94].
Table 7.1 shows the tabulated property values for each region of the single and
multiple inclusion simulation cases, including their positions and dimensions in the X-Y
plane. The FEM mesh used to calculate the forward solution for each case is shown in
figure 7.1. The results from this investigation will be discussed in the following section.
For each of these 2D experiments, the error has been calculated as:
∑
√(
()
( ))
⁄
146
()
(7.1)
where n is the number of nodes in the region of interest, Meas(i) is the property at the i th
node and Exact (i) is the exact probe measurements at the ith node, respectively.
Additionally, the error has been normalized to the number of nodes in the reconstruction
mesh.
Table 7.1
Table 7.1. Position and property values of the regions for the three simulation cases
(shown in figure 7.1).
(a)
(b)
(c)
Figure 7.1. Simulation meshes used for calculating the forward data using a FEM
method described in [##]: (a) single, (b) two and (c) three inclusion cases showing
the background (dark blue), breast (light blue) and inclusion(s) (green, orange and
red) regions, respectively.
147
Case 1: A single inclusion
Reconstructed permittivity and conductivity images for the single inclusion case
are shown in figure 7.2 and 7.3, respectively. The corresponding inclusion property
values and the % error over the breast region are tabulated in table 7.2.
Table 7.2
Table 7.2. Recovered permittivity and conductivity values for the one inclusion
case, including error calculations for the uniform (top) and conformal (bottom)
images, respectively.
(a)
(b)
Figure 7.2. Reconstructed (a) conformal and (b) uniform permittivity images of the
simulation set-up shown in figure 7.1a.
148
(a)
(b)
Figure 7.3. Reconstructed (a) conformal and (b) uniform conductivity images of the
simulation set-up shown in figure 7.1a.
Results from this simulation investigation show improved property recovery for
the inclusion’s permittivity and conductivity values as well as the corresponding error
over the breast region. An increase of 11% and 4% for the maximum permittivity and
conductivity values is seen when using the CMI reconstruction process compared to the
uniform technique. Additionally, the error over the breast region decreased by 2% for
both the permittivity and conductivity values by using a boundary conformed
reconstruction mesh.
This error reduction is due to limiting the smoothing that occurs at dielectrically
contrasting region interfaces due to our reconstruction algorithm’s least squares
minimization approach; by conforming the reconstruction mesh to the OUT’s boundary
we have reduced the smoothing by eliminating the dielectrically contrasting region that
surrounds the OUT. For each of the inclusion cases (1, 2, and 3), the error over the breast
region remained relatively constant for both the permittivity and the conductivity
reconstructions (2% breast error reduction), and consequently the two and three inclusion
cases will specifically focus on the enhancement of the inclusion regions due to the CMI
procedure.
149
Case 2: Two inclusions
Reconstructed permittivity and conductivity images for the two-inclusion case are
shown in figure 7.4 and 7.5, respectively. The corresponding inclusion’s property values
are tabulated in tables 7.3 (green-inclusion from figure 7.1b) and 7.4 (orange-inclusion
from figure 7.1b).
Table 7.3
Table 7.3. Recovered permittivity and conductivity values for the 1st inclusion of
the two-inclusion case (green-inclusion in figure 7.1b).
Table 7.4
Table 7.4. Recovered permittivity and conductivity values for the 2nd inclusion of the
two-inclusion case (orange-inclusion in figure 7.1b).
150
(a)
(b)
Figure 7.4. Reconstructed (a) conformal and (b) uniform permittivity images of the
simulation set-up shown in figure 7.1b.
(a)
(b)
Figure 7.5. Reconstructed (a) conformal and (b) uniform conductivity images of the
simulation set-up shown in figure 7.1b.
Results from this simulation investigation show improved property recovery for
the permittivity and conductivity values for both inclusions using the CMI procedure.
Using this technique, inclusion region one experiences an increase of 2% and 3.9% for
the permittivity and conductivity values, respectively. The 2nd inclusion region’s
permittivity and conductivity were enhanced by 7% and 5.9%, respectively. The
improved detection of inclusion two is presumably due to its location within the
simulated set-up. We have shown that our reconstruction algorithm has increased
151
sensitivity regarding the recovery of high contrast regions that are located closer to the
antenna elements, which may have an additional impact on the detection of inclusion one
whose improvements were more modest compared to the single inclusion case results.
152
Case 3: Three inclusions
Reconstructed permittivity and conductivity images for the three-inclusion case
are shown in figure 7.6 and 7.7, respectively. The corresponding inclusion property
values are tabulated in tables 7.5 (green-inclusion from figure 7.1c), 7.6 (orangeinclusion from figure 7.1c) and 7.7 (red-inclusion from figure 7.1c).
Table 7.5
Table 7.5. Recovered permittivity and conductivity values for the 1st inclusion of the
three-inclusion case (green-inclusion in figure 7.1c).
Table 7.6
Table 7.6. Recovered permittivity and conductivity values for the 2nd inclusion of the
three-inclusion case (orange-inclusion in figure 7.1c).
153
Table 7.7
Table 7.7. Recovered permittivity and conductivity values for the 3rd inclusion of
the three-inclusion case (red-inclusion in figure 7.1c).
(a)
(b)
Figure 7.6. Reconstructed (a) conformal and (b) uniform permittivity images of the
simulation set-up shown in figure 7.1c.
(a)
(b)
Figure 7.7. Reconstructed (a) conformal and (b) uniform conductivity images of the
simulation set-up shown in figure 7.1c.
154
Results from this simulation investigation show improved property recovery for
the permittivity and conductivity values for all inclusions using the CMI reconstruction
process. Inclusion one experienced a 16% and 5% increase in its permittivity and
conductivity values, respectively. For the 2nd inclusion region, the permittivity and
conductivity increased by 14% and 1.5%, respectively. The third inclusion experienced
permittivity and conductivity increases of 10.6% and 2%, respectively.
Results from these 2D simulations show similar trends regarding the use of CMI
reconstruction; by conforming the property mesh to the size of the OUT’s boundary,
property recover was enhanced for both the permittivity and conductivity. Especially
interesting is the large permittivity increase of the 3rd inclusion in case three. Its radius
was ½ the size of the other inclusions used in that set-up. This seems to indicate that the
use of CMI reconstruction significantly enhances the detection of physically smaller
dielectrically contrasting regions within the OUT.
155
7.2. Phantom Studies
A 2D phantom experiment has been conducted to confirm the efficacy of using
the CMI procedure with our imaging system. When using the conformal meshing
technique, it is necessary to separate the imaging domain into two distinct regions: the
coupling-medium and the OUT, shown as region 1 and region 2 in figure 7.8a. In order to
produce accurate information regarding the OUT, all nodes in the conformed mesh
should be contained on its boundary, with no nodes extending into the coupling medium
zone. Adhering to this fundamental requirement allows the reconstruction mesh to be
constructed in any shape at any location within the antenna array. Additionally, an
assessment of the recovered permittivity as a function of the mismatch between the
reconstruction mesh and the actual OUT’s boundary will be presented to verify optimal
mesh scaling.
(a)
(b)
Figure 7.8. (a) Schematic of the experimental imaging domain and (b) a photograph of
the phantom placed in the imaging system.
156
For this analysis, a cylindrical phantom (86:14 glycerine:water) with a submerged
inclusion (60:40 glycerine:water) was imaged using the current MI system. A schematic
of the imaging domain and a photograph of the phantom placed in the imaging system are
shown in figure 7.8a and 7.8b, respectively. The corresponding DPs of the phantom
materials and their position in the X-Y plane are shown in table 7.8.
Table 7.8.
Table 7.8. The DPs and positions of the phantom materials shown in figure 7.7b.
(a)
(b)
Figure 7.9. Reconstructed permittivity images using a (a) uniform 7 cm and (b)
exact conformal 5.5 cm radius reconstruction mesh, respectively.
157
Figure 7.9 shows the reconstructed phantom images at 1300 MHz; results from
this investigation show a 10.5% increase in the recovered permittivity values of the tumor
region and a 4.75% decrease in the error over the breast region [125]. The permittivity
values and the % error over the breast region are tabulated in table 7.9.
Table 7.9
Table 7.9. Recovered permittivity values and the corresponding % error over the
breast region for the phantom shown in figure 7.8.
158
Parameter mesh scaling and translation
As indicated in [124] and [125], the scaling of the parameter mesh can impact the
resulting image accuracy. Figure 7.10 shows the reconstructed permittivity images for the
phantom shown in figure 7.8; each image in the series utilized a different sized parameter
mesh for image reconstruction, with radii ranging from 7 cm to 4.5 cm. This mesh scaling
experiment allowed an analysis of image quality as a function of mismatch between the
computed zone and the target region. The recovered permittivity values along with the
corresponding error values are summarized in table 7.10.
Table 7.10
Table 7.10. Position, permittivity values and error calculations of the phantom shown
in figure 7.7 for a variety of parameter meshes (7 cm represents the uniform mesh and
5.5 cm represent the exact boundary conformed mesh, respectively).
159
Figure 7.10. Reconstructed permittivity images of the phantom shown in figure 7.7
for a variety of parameter meshes: (a) 7 cm, (b) 6 cm, (c) 5.5 cm (exact) (d) 5 cm and
(e) 4.5 cm radii, respectively.
160
From these images, an OUT-parameter mesh mismatch metric has been extracted,
and is shown in figure 7.11. As indicated from the plot, the exact permittivity (red-line)
corresponds to the parameter mesh that has zero boundary-mismatch (the exact boundary
conformed mesh). This finding shows that the optimal DP value (in this case permittivity)
is achieved by using a reconstruction mesh that is conformed to OUT’s perimeter. An
additional interesting finding is seen by inspection figure 7.9d and 7.9e. As the property
mesh is scaled smaller than the OUT’s boundary, low property artifacts appear; this has
additionally been identified in [118,125].
Figure 7.11. Plot of the inclusion’s permittivity as a function of OUT-parameter
mesh mismatch.
Figure 7.12 shows the conformally reconstructed image of the phantom in figure
7.8 for a variety of horizontal parameter mesh translations. The mismatch between the
target zone and the optimal mesh position show similar trends as the mesh scaling
analysis. For these reconstructions, the same phantom (shown in figure 7.8) was imaged
161
with the inclusion located at X-Y position (2.5, 2). By inspecting the images it is
observed that increasing position mismatch is associated with larger low-property
artifacts which occur on the complimentary side relative to the horizontal translations.
Although slight variations in the inclusion’s permittivity values are seen, the appearance
of the low property values is evidence of target zone-mesh position mismatch; with larger
mismatches resulting in enhanced low property artifacts, which are undesired. This
finding is critical; the results show that even a 5 mm mismatch in mesh placement can
degrade image quality. The ability to accurately obtain the OUT’s boundary is now
possible with the use of the mounted optical scanner. It will be discussed in more detail in
section 7.4 preceded by 3D simulation experiments that show similar trends to the 2D
position mismatch effects, e.g. the appearance of low property artifacts.
Figure 7.12. Reconstructed permittivity images of the phantom shown in figure 7.7
for a variety of horizontal translations (top) +5 to +25 mm translations in 5 mm
increments, middle exact conformal and (bottom) -5 to -25mm translations in 5 mm
increments.
162
7.3. 3D Simulation Studies
A series of 3D simulations has been conducted to analyze the effect of using the
CMI process in conjunction with our 3D reconstruction algorithm. We have already seen
that horizontal mismatch results in the generation of artificial features. This experiment
will analyze the effect of mismatch in the vertical direction as well. The results show
similar trend as the 2D case; conforming the reconstruction mesh to the OUT boundary
enhanced DP recovery. The positions and properties of the three regions in this 3D
simulation are tabulated in table 7.11. Figure 7.13 shows a schematic of the simulated
imaging domain. The reconstructed conductivity images (1300 MHz) for a variety of
Figure 7.13. Schematic of the imaging domain used for the 3D simulations shown in
figure 7.14 including a bath (light blue), breast (dark blue) and tumor (red) regions.
Table 7.11
Table 7.11. Position and conductivity properties of the simulated imaging domain
shown in figure 7.13.
163
parameter mesh heights and radii are shown in figure 7.14. They have been iso-surfaced
to highlight the mismatches effect on the recovery of the tumor-like region.
(a)
(b)
(c)
(d)
Figure 7.14. 1300 MHz reconstructed conductivity images (from simulated data
corresponding to the imaging domain shown in figure 7.12) using varying mesh
heights and radii: (a) 7 cm radius with a 4 cm height, (b) 5.75 cm radius with a 3.5
cm height, (c) 5.0 cm radius with a 3.25 cm height (exact) and (d) 4 cm radius
with a 3 cm height.
The scaling effects seen in figure 7.14 are consistent with the 2D analysis, optimal
inclusion recovery is achieved by utilizing a boundary conformed mesh in the
reconstruction process. Additionally apparent is the low property manifestations that
appear if the reconstruction’s mesh is smaller than the target zone (figure 7.14d). As a
result of these mismatch investigations we know that even a 5 mm mesh position
mismatch can result in the generation of artifacts in the reconstructed images.
Consequently, the scanner has been designed with sub-mm accuracy, allowing accurate
detection of the OUT while submerged inside out system’s illumination chamber.
164
7.4. Surface Mounted Laser Scanning System Studies
With the use of the optical scanner, we can now extract the boundary of an OUT
while submerged in the imaging system. Figure 7.15 shows a 1 mm thick plastic breast
phantom placed in the system’s illumination chamber. The phantom was filled with an
86:14 glycerin:water solution and contained a gelatin inclusion made of an 15:85
gelatin:water solution. Using a calibration procedure to account for any distortion at the
tank walls, the scanner can measure and extract the boundary of the OUT (in this case the
breast phantom). Figure 7.16 shows the resulting point-cloud and wire surface mesh to
create the custom OUT mesh for image reconstruction.
(a)
(b)
Figure 7.15. Photographs of (a) plastic breast mold (b) submerged in the imaging
system’s illumination chamber.
165
(a)
(b)
Figure 7.16. Scanner extracted (a) point cloud and (b) wire mesh of the breast
phantom shown in figure 7.15.
From the point cloud, a custom volumetric OUT mesh can be generated and used
in the CMI reconstruction process. Figure 7.17 shows reconstructed permittivity and
conductivity images of the breast phantom using the custom scanner-generated mesh. The
corresponding DPs are tabulated in table 7.12. For comparison purposes, the uniform
mesh reconstructions are shown in figure 7.18.
Table 7.12.
Table 7.12. Dielectric properties of the breast phantom and tumor inclusion at
1300 MHz.
166
(a)
(b)
Figure 7.17. CMI reconstructions (1300 MHz) of the breast phantom shown in figure
7.15 (a) permittivity and (b) conductivity.
(a)
(b)
Figure 7.18. Uniform mesh reconstruction utilizing the same data as the
reconstructed images in figure 7.17.
167
The use of the optical scanner significantly increased DP accuracy and tumor
location. Its use has resulted in a 15% increase of the recovered tumor’s permittivity and
conductivity values. Although the uniform mesh reconstruction technique is more than
capable of detecting dielectrically contrasting regions within the breast-like phantom, the
use of the scanner-generated reconstruction mesh has significantly increased the accuracy
of dielectrically contrasting regions within the OUT.
168
7.5 Conclusion
3D conformal microwave image reconstruction is now possible. The analysis in
this chapter has shown that its use increases the accuracy of the resulting DP values for
both 2D and 3D image reconstruction. This has significant clinical implications;
extracting the breast boundary while submerged in the illumination chamber reduces the
need to utilize external modalities (MRI) for obtaining the requisite boundary
information. This reduces image registration issues associated with using more than one
modality to collect all needed information for image reconstruction. And as shown in this
chapter’s analysis, its use enhances DP recovery (permittivity and conductivity) in 2D
and 3D. Combined with the optimized motion control system, we are now capable of
administering 3D patient exams in clinically feasible time frames while collecting
optimal data and breast boundary information for enhanced image quality.
169
8. Summary
The goals outlined for the work in this thesis have been fulfilled, resulting in the
development of an advanced MI prototype with increased data acquisition capabilities.
Due to microwave electronic and motion control hardware and software advances, the
current system can acquire optimal data required for imagine reconstruction while
obtaining OUT boundary information used to enhance the resulting images. 3D data
acquisition speeds have increased, resulting in faster 3D patient examinations. Although
previously limited due to lengthy data acquisition times associated with non-optimized
data collection, these hardware and software improvements have decreasing 3D exam
times. This is a significant clinical achievement that positively impacts the patient and the
system operator. The results from the experiments in this work have identified ideal
frequency and SA spacing metrics which permit the collection of optimal data for image
reconstruction with reduced time requirements. The system’s sensitivity to signal power
has been thoroughly examined. Limited error is seen while operating up to 2 GHz, with
performance degrading at higher operating frequencies. Isolation issues have been
identified and successful solutions have been presented, allowing the retention of a
streamlined electronic configuration as well as system modularity.
170
References
1. L. E. Larsen, and J. H. Jacobi, Medical Applications of Microwave Imaging, IEEE,
New York, 1986.
2. S. M. Michaelson, and J. C. Lin, Biological Effects and Health Implications of
Radiofrequency Radiation, New York, NY: Plenum, 1987.
3. J. Thuery, Microwaves: Industrial, Scientific and Medical Applications, Boston, MA:
Artech House, 1992.
4. L. A. Geddes, and L. E. Baker, Principles of Applied Biomedical Instrumentation, 3rd
edition, New York, NY: Wiley, 1989.
5. A. V Vorst, A. Rosen, and Y. Kotsuka, RF/Microwave Interaction with Biological
Tissue, New York, NY: Wiley, 2006.
6. H. P. Schwan, “Electrical properties of tissue and cell suspensions,” Advanced
Physics in Medicine and Biology, vol. 5, pp.147-209, 1957.
7. F. A. Duck, Physical Properties of Tissue: A Comprehensive Reference Book,
London, UK: Academic Press, 1990.
8. E. C. Burdette, F. L. Cain, and J. Seal, “In situ permittivity measurements:
perspectives, techniques, results,” Medical Application of Microwave Imaging, IEEE
Press 1986.
169
9. M. Lazebnik, D. Popovic, L. McCartney, C. B. Watkins, M. J. Lindstrom, J. Harter,
S. Sewall, T. Ogilvie, A. Magliocco, T. M. Breslin, W. Temple, D. Mew, J. H.
Booske, M. Okoniewski, and S. C. Hagness, "A large-scale study of the ultrawideband microwave dielectric properties of normal, benign and malignant breast
tissues obtained from cancer surgeries," Physics in Medicine and Biology, vol. 52, pp.
6093-115, Oct 21 2007.
10. K. Carr, “Microwave radiometry: its importance to the detection of cancer,” IEEE
Transactions on Microwave Theory and Techniques, vol. 37, pp. 1862-69, 1989.
11. B Bocquet, R Ait-Abdelmalek, A Mamouni, K Ridaoui, and Y Leroy, “Influence of
the permittivity of breast tissue on the robust of microwave radiometric,” IEEE
Applications of Microwaves in Medicine, vol. 6, pp. 1-5, 1995.
12. A. H. Barrett, P. C. Myers, and N. L. Sadowsky, “Detection of breast cancer by
microwave radiometry,’’ Radiology in Science, vol. 12, pp. 167, 1977.
13. P. C. Myers, N. L. Sadowsky, and A. H. Barrett, “Microwave thermograph:
principles, methods and clinical applications.” Journal of Microwave Power, vol. 14,
1979.
14. J. Edrich. “Centimeter- and millimeter-wave thermography- a survey on tumor
detection.” Journal of Microwave Power, vol. 14, pp. 95-104. 1979.
15. B. Enander and G. Larson, “Microwave radiometric measurements on the
temperature inside the body,” Electronic Letter, vol. 10, pp. 317-319, 1974.
172
16. A. Mamouni, F. Bliot. Y. Leroy, and Y. Moschetto, “A modified radiometer for
temperature and microwave properties measurements of biological substances,” in 7th
European Microwave Conference, (Copenhagen, Denmark), pp. 703-707, 1977.
17. K. L. Carr, A. M. El-Mahdi, and J. Shaeffer, “Dual-mode microwave system to
enhance early detection of cancer,” IEEE Transactions on Microwave Theory and
Technology, vol. 29, pp. 256-260, 1981.
18. R. A. Porter, and H. H. Miller, “Microwave radiometric detection and location of
breast cancer,” Boston, MA: ELECTRO, 1978.
19. F. Bardati, and S. Ludicello, “Modeling the visibility of breast malignancy by a
microwave radiometer,” IEEE Transactions on Biomedical Engineering, vol. 55, no.
1, pp. 214-21, 2008.
20. R. U, K. Noell, K. Woodward, B. Worde, R. Fishbum, and L. Miller, “Microwaveinduced local hyperthermia in combination with radiotherapy of human malignant
tumors,” Cancer, vol. 45, pp. 638-46, 1980.
21. R. N. Lawson and M. S. Chughtai, “Breast cancer and body temperature,” Canadian
Medical Association Journal, vol. 88, pp. 68-70, 1963.
22. K. R. Foster, and E. A. Cheever, "Microwave radiometry in biomedicine - a
reappraisal," Bioelectromagnetics, vol. 13, pp. 567-579, 1992.
173
23. G. Giaux, J. Delannoy, D. Delvalee, Y. Leroy, A. Mamouni, J. C. Vandevelde, and B.
Bocquet, "Microwave radiometry imaging - characterization of breast-tumors,"
Journal of Photographic Science, vol. 39, pp. 164-65, 1991.
24. H. Hinrikus, J. Riipulk, and H. Podra, "Design of microwave radiometer for early
detection of cancer," Physical Medicine, vol. 13, pp. 324-325, 1997.
25. M. H. Xu, "Microwave-induced thermoacoustic tomography using multi-sector
scanning," Medical Physics, vol. 28, pp. 1958-63, 2001.
26. B. Guo, J. Li, H. Zmuda, and M. Sheplak, "Multifrequency microwave-induced
thermal acoustic imaging for breast cancer detection," IEEE Transactions on
Biomedical Engineering, vol. 54, pp. 2000-10, 2007.
27. G. Ku, and L. V. Wang, "Scanning microwave-induced thermoacoustic tomography:
signal, resolution, and contrast," Medical Physics, vol. 28, pp. 4-10, 2001.
28. J. P. Reilly, L. A. Geddes, and C. Polk, Bioelectricity: The Electrical Engineering
Handbook, Boca Raton, FL: CRC Press, 1993.
29. E. G. Moros, J. Penagaricano, P. Novak, W. L. Straube, and R. J. Myerson, "Present
and future technology for simultaneous superficial thermoradiotherapy of breast
cancer," International Journal of Hyperthermia, vol. 26, pp. 699-709, 2010.
30. P. M. Meaney, M. W. Fanning, D. Li, S. P. Poplack, and K. D. Paulsen, "A clinical
prototype for active microwave imaging of the breast," IEEE Transactions on
Microwave Theory and Techniques, vol. 48, pp. 1841-53, 2000.
174
31. E. C. Fear, and M. A. Stuchly, "Microwave system for breast tumor detection," IEEE
Microwave and Guided Wave Letters, vol. 9, pp. 470-72, 1999.
32. P. M. Meaney, K. D. Paulsen, A. Hartov, and R. K. Crane, "An active microwave
imaging-system for reconstruction of 2-D electrical property distributions," IEEE
Transactions on Biomedical Engineering, vol. 42, pp. 1017-26, 1995.
33. S. Caorsi, G. L. Gragnani, M. Pastorino, and M. Rebagliati, "An approach to focused
2D tm near-field active microwave imaging of dielectric objects," Microwave and
Optical Technology Letters, vol. 8, pp. 302-6, 1995.
34. V. Zhurbenko, T. Rubaek, V. Krozer, and P. Meincke, "Design and realisation of a
microwave three-dimensional imaging system with application to breast-cancer
detection," Institute of Engineering and Technology Microwaves Antennas &
Propagation, vol. 4, pp. 2200-211, 2010.
35. A. Santorelli, E. Porter, J. Schwartz, and M. Popovi , “Pulse Shaping for timedomain microwave breast tumor detection: experiments with realistic tissue
phantoms,” IEEE European Conference on Antennas and Propagation, pp. 939-41,
2011.
36. J. Bourqui, J. M. Sill, and E. C. Fear, “A prototype system for measuring microwave
frequency reflections from the breast,” International Journal of Biomedical Imaging,
vol. 2012, article ID 851234, 2012.
175
37. M. Klemm, I. Craddock, J. Leendertz, A. Preece, and R. Benjamin, "Experimental
and clinical results of breast cancer detection using UWB microwave radar." IEEE
Antennas and Propagation Symposium, pp. 1-4, 2008.
38. M. Klemm, J. Leendertz, D. Gibbins, I. J. Craddock, A. Preece, and R. Benjamin,
“Microwave radar-based differential breast cancer imaging: imaging in homogenous
breast phantoms and low contrast scenarios,” IEEE Transactions on Antennas and
Propagation, vol. 58, pp. 2237-344, 2010.
39. Q. Fang, P. M. Meaney, and K. D. Paulsen, “Microwave imagine reconstruction of
tissue property dispersion characteristics utilizing multi-frequency information,”
IEEE Transactions on Microwave Theory and Techniques, vol. 53, pp. 1866-875,
2004.
40. X. Zeng, A. Fhager, P. Linner, M. Persson, and H. Zirath, “Experimental
investigation of the accuracy of an ultra-wideband time-domain microwavetomographic system,” IEEE Transactions on Instrumentation and Measurements, vol.
60, pp. 3939-949, 2011.
41. D. Li, P. M. Meaney, T. Raynolds, S, A, Pendergrass, M. W. Fanning, an K. D.
Paulsen, “Parallel-detection microwave spectroscopy system for breast imaging,”
Review of Scientific Instruments, vol. 75, pp. 2305-313, 2004.
42. K. Kwon, Y. Lim, C. Kim, N. Kim, C. Park, K. Yoo, s. Son and S. Jeon, “Microwave
tomography analysis system for breast tumor detection,” Journal of Medical Systems,
vol. 36. Pp. 1757-767, 2012.
176
43. A. Salvade`, M. Pastorino, R. Monleone, G. Bozza and A. Randazzo, “A new
microwave axial tomograph for the inspection of dielectric materials,” IEEE
Transactions on Instrumentation and Measurements, vol. 58. pp. 2072-79, 2009.
44. P. M. Meaney, T. Grzegorczy, E. Attardo, and K. D. Paulsen, “Ultrafast 3D
microwave
tomography
utilizing
the
direct
dipole
approximation,”
IEEE
Electromagnetics in Advanced Applications, pp. 838-39, 2012.
45. P. M. Meaney, “Microwave imaging and emerging applications,” International
Journal of Biomedical Imaging, vol. 2012, Article ID 252093, 2012.
46. C. H. Yip, E. Cazap, B. O. Anderson, L. J. Passman, L. A. Santint, R. C. da Silva, D.
B. Thomas, Y. Zheng, and H. M. Khaled, “Breast cancer management in middleresource countries: consensus statement from the breast health global initiative,”
Breast, vol. 20, 2011.
47. “U.S.
breast
cancer
statistics.”
Internet:
http://www.breastcancer.org/symptoms/understand_bc/statistics.jsp, Mar. 1, 2012
[Oct. 19, 2011].
48. American Cancer Society Cancer Facts and Figures 2012. Atlanta, GA: American
Cancer Society, 2012.
49. Ronckers CM, and Erdmann CA, “Radiation and breast cancer: a review of current
evidence,” Breast Cancer Research, vol. 7, pp. 21-32, 2005.
177
50. P.E. Goss, and S. Sierra, “Current perspectives on radiation-induced breast cancer,”
Journal of Clinical Oncology, vol. 16, pp. 338-47, 1998.
51. P. H. Zahl, B. H. Strand, J. Maehlen, “Incidence of breast cancer in Norway and
Sweden during introduction of nationwide screening: prospective cohort study,”
Journal of Biomedical Imaging, vol. 328, pp. 921-4, 2004.
52. A. Bleyer, H. G. Welch, “Effect of three decades of screening mammography on
breast-cancer incidence,” The New England Journal of Medicine, vol. 367, pp.19992005, 2012.
53. J. G. Elmore, M. B. Barton, and V. M. Moceri, “Ten-year risk of false positive
screening mammograms and clinical breast examinations,” The New England Journal
of Medicine, vol. 338, pp. 1089-96, 1998.
54. R. D. Rosenberg, W. C. Hunt, and M. R. Williamson, “Effects of age, breast density,
ethnicity, and estrogen replacement therapy on screening mammographic sensitivity
and cancer stage at diagnosis review of 183,134 screening mammograms in
Albuquerque, New Mexico,” Radiology, vol. 209, pp. 511-8, 1998.
55. K. Kerlikowske, D. Grady, and J. Barclay, “Likelihood ratios for modern screening
mammography: risk of breast cancer based on age and mammographic
interpretation,” Journal of the American Medical Association, vol. 276, pp. 39-43,
1996.
178
56. P. L. Porter, A Y. El-Bastawissi, and M. T. Mandelson, “Breast tumor characteristics
as predictors of mammographic detection: comparison of interval- and screendetected cancers,” Journal of the National Cancer Institute, vol. 91, pp. 2020-8, 1999.
57. National
Institute
of
Biomedical
Imaging
and
Bioengineering:
Internet:
http://www.nibib.nih.gov/NewsEvents/ResearchHighlights/HighlightOnHealth/30Au
g12. [Sept. 19, 2012].
58. W. T. Joines, Y. Zhang, C. Li, and R. L. Jirtle, "The measured electrical properties of
normal and malignant human tissues from 50 to 900 MHz," Medical Physics, vol. 21,
pp. 547-50, Apr 1994.
59. C. Polk, and E. Pastow, Handbook of Biological Effects of Electromagnetic Fields,
Boca Roton, FL: CRC Press, 1996.
60. S. R. Smith, and K. R. Foster, "Dielectric-Properties of Low-Water-Content Tissues,"
Physics in Medicine and Biology, vol. 30, pp. 965-973, 1985.
61. H. Schwan, and K. Foster, “RF-field interactions with biological systems: electrical
properties and biophysical mechanisms,” Proceedings of the IEEE, vol. 68, pp.10413, 1980.
62. G. Bindu, Dielectric Behavior of Biological Objects, Boca Raton, FL: CRC Press,
1993.
179
63. L. M. Oliver, E. T. McAdams, P. S. Dunlop, J, A, Byrne, I. S. Blair, M. Boyle, and K.
G. McGuigan, Electrical Impedance Spectroscopy as a Powerful Analytic Tool for
Monitoring Microbial Growth on Medical Implants, Encyclopedia of Healthcare
Information Systems, 2008.
64. S. Grimnes, and O. G. Martinsen, Bioimpedance and Bioelectricity Basics, San
Diego, CA: Academic Press, 2000.
65. W. Kuang and S. Nelson, “Low-frequency dielectric properties of biological tissues: a
review with some new insights,” Transactions of the American Society of
Agricultural Engineers, vol. 41, pp. 173-84, 1998.
66. G. Grosse, “Permittivity of a suspension of charged spherical particles in electrolyte
solution. II Influence of the surface conductivity and asymmetry of the electrolyte on
the low and high frequency relaxations,” Journal of Physical Chemistry, vol. 92, pp.
3905-910, 1988.
67. G. Schwarz, “A theory of the low frequency dielectric dispersion of colloidal
particles in electrolyte solution,” Journal of Physical Chemistry, vol. 66, pp. 2636,
1962.
68. G. Schwarz, “Uber die dispersion des orientierungsfieldeffektes von polyelektrolten
in hochfrequenten elektrischen feldern,” Journal of Physical Chemistry (Frankfurt),
vol. 19, pp. 286, 1959.
180
69. C. Gross, and K. Foster, “Permittivity of a suspension of charged spherical particles
in electrolyte solutions,” Journal of Physical Chemistry, vol. 91, pp. 3073, 1987.
70. H. Schwan, Electrical properties of cells: principled, some resent results and some
unresolved problems, the biological approach to excitable systems, New York, NY:
Plenum, 1981.
71. K. Foster, and S. Schwan, “Dielectric properties of tissue” in CRC Handbook of
Biological Effects of Electromagnetic Fields, Boca Raton, FL: CRC Press, 1996.
72. K. Foster, and J. Schepps, “Dielectric properties of normal and malignant tissues at
radio through microwave frequencies,” Journal of Microwave Power, vol. 16. Pp.
107-19, 1981.
73. R. Pethig, “ Dielectric properties of biological materials: biophysical and medial
applications,” IEEE Transactions on Electrical Insulation, vol. 19, pp. 453-72, 1984,
74. S. R. Smith, and K. Foster, “Dielectric properties of low-water-content tissues,”
Physics in Medicine and Biology, vol. 30, pp. 369, 1984.
75. D. Misra, M. Channra, B. R. Epstein, M. Mirotznik, and K. Foster, “Noninvasive
electrical characterization of materials at microwave frequencies using open-ended
coaxial line: test of an improved calibration technique,” IEEE Transactions on
Microwave Theory and Techniques, vol. 38. Pp. 8-14, 1990.
181
76. F. Duhamel, I. Huynen, and A. Vander Vorst, “Measurement of complex permittivity
of biological and organic liquids up to 110 GHz,” IEEE MTT-S Microwave
International Symposium Digest, Denver, CO, pp.107-110, 1997.
77. S. Gabriel, R. Lau, and C. Gabriel, “The dielectric properties of biological tissues: II.
Measurements in the frequency range 10 Hz to 20 GHz,” Physics in medicine and
biology, vol. 41, no. 11, pp. 2251–2270, 1996.
78. H. Woodard and D. White, “The composition of body tissues,” British Journal of
Radiology, vol. 59, no. 708, p. 1209, 1986.
79. T. Zhou, “Microwave tomographic imaging using dielectric contrast in temperature
and tissue content,” PhD, Thayer School of Engineering, Dartmouth College,
Hanover, NH, 2011.
80. J. Schepps and K. Foster, “The UHF and microwave dielectric properties of normal
and tumor tissues: variation in dielectric properties with tissue water content,”
Physics in Medicine and Biology, vol. 25, pp. 1149–1159, 1980.
81. F. Homburger and W. Fishman, The physiopathology of cancer: a treatise for
investigators, physicians, and students, p. 554. Hoeber-Harper, 1953.
82. J. Bushberg, J. Seibert, E. Leidholdt Jr, J. Boone, and E. Goldschmidt Jr, “The
essential physics of medical imaging,” Medical Physics, vol. 30, p. 1936, 2003.
83. J. Greenstein, Biochemistry of Cancer, p. 246. Academic Press, Inc. New York, 1947.
182
84. M. Lazebnik, D. Popovic, L. McCartney, C. B. Watkins, M. J. Lindstrom, J. Harter,
S. Sewall, T. Ogilvie, A. Magliocco, T. M. Breslin, et al., “A large-scale study of the
ultra-wideband microwave dielectric properties of normal, benign and malignant
breast tissues obtained from cancer surgeries,” Physics in medicine and biology, vol.
52, no. 20, pp. 6093–6116, 2007.
85. L. Sha, E. R. Ward, and B. Story, “A review of dielectric properties of normal and
malignant breast tissue,” IEEE Proceedings Southwest Conference, pp. 457, 2002.
86. V. McCormack, and I. Silva, “Cancer epidemiology, biomarkers and prevention,”
American Association for Cancer Research, vol. 15, pp. 1159-169, 2006.
87. X. Li, E. J. Bond, B. D. Van Veen, and S. C. Hagness, “An overview of ultrawideband microwave imaging via space-time beamforming for early-stage breastcancer detection,” IEEE Antennas and Propagation Magazine, vol. 47, pp. 19-34,
2005.
88. S. Poplack, T. Tosteson, W. Wells, B. Pogue, P. Meaney, A. Hartov, C. Kogel, S.
Soho, J. Gibson, and K. Paulsen, “Electromagnetic breast imaging: results of a pilot
study in women with abnormal mammograms,” Radiology, vol. 243, no. 2, p. 350,
2007.
89. P. M. Meaney, P. A. Kaufman, L. S. Muffly, M. Click, S. P. Poplack, W. A. Wells,
G. N. Schwartz, R. M. di Florio-Alexander, T. D. Tosteson, Z. Li, S. D. Geimer, M.
W. Fanning, T. Zhou, N. R. Epstein, and K. D. Paulsen, “Microwave imaging for
183
neoadjuvant chemotherapy monitoring: initial clinical experience,” Breast Cancer
Research, (accepted manuscript ID 9777187567643131), 2013.
90. E. Tanabe and W.T. Joines, “A nondestructive method for measuring the complex
permittivity of dielectric materials at microwave frequencies using an open
transmission
line
resonator,”
IEEE
Transactions
On
Instrumentation
and
Measurements, vol. 25, pp. 222-226, 1976.
91. A. Stogryn, “Equations for calculating the dielectric constant of saline water,” IEEE
Transactions On Microwave Theory and Techniques, vol. 19, pp. 733-736, 1971.
92. E. C. Fear, S. C. Hagness, and P. M. Meaney, “Enhancing breast tumor detection with
near-field imaging,” IEEE Microwave magazine, vol. 3, pp. 48-56, 2002.
93. P. M. Meaney, K. D. Paulsen, and T. P. Ryan, “2-Dimensional hybrid element Imagereconstruction for tm illumination,” IEEE Transactions on Antennas and
Propagation, vol. 43, no. 3, pp. 239-247, Mar, 1995.
94. P. M. Meaney, Q. Fang, T. Rubaek, E. Demidenko, and K. D. Paulsen, "Log
transformation benefits parameter estimation in microwave tomographic imaging,"
Medical Physics, vol. 34, pp. 2014-2023, 2007.
95. Q. Fang, P. M. Meaney, K. D. Paulsen, "Viable three-dimensional microwave
imaging: theory and experiments," IEEE Transactions on Antennas and Propagation,
vol. 58, pp. 449-458, 2010.
184
96. A. Fhager, P. Hashemzadeh, and M. Persson, “Reconstruction quality and spectral
content of an electromagnetic time-domain inversion algorithm,” IEEE Transactions
on Biomedical Engineering, vol. 53, pp. 1594-1604, 2006.
97. P. M. Meaney, M. W. Fanning, T. Raynolds, C. J. Fox, Q. Fang, C. Kogel, S. P.
Poplack, and K. D. Paulsen, “Initial clinical experience with microwave breast
imaging in women with normal mammography,” Academic Radiology, vol. 10, pp.
207-218, 2006.
98. P. M. Meaney, T. Raynolds, L. Potwin, and K. D. Paulsen, "3-point support
mechanical steering system for high intensity focused ultrasound," Physics in
Medicine and Biology, vol. 52, pp. 3045-3056, 2007.
99. A. Golnabi, “Computational aspects of tomographic microwave imaging for
biomedical applications,” PhD, Thayer School of Engineering, Dartmouth College,
Hanover, NH, 2012.
100.
P. M. Meaney, M. W. Fanning, D. Li et al., “A clinical prototype for active
microwave imaging of the breast,” IEEE Transactions on Microwave Theory and
Techniques, vol. 48, no. 11, pp. 1841- 1853, 2000.
101.
P. M. Meaney, A. H. Golnabi, N. R. Epstein, S. D. Geimer, M. W. Fanning, J. B.
Weaver, and K. D. Paulsen, “Integration of microwave tomography with magnetic
resonance for improved breast imaging,” Medical Physics (submitted).
185
102.
D. M. Pozar, Microwave and RF Design of Wireless Systems, New York, NY:
Wiley, 2000.
103.
Pulsar Microwave Corporation: http://www.pulsarmicrowave.com/.
104.
C. J. Fox, P. M. Meaney, F. Shubitidze, L. Potwin, and K. D. Paulsen,
“Characterization of an implicitly resistively-loaded monopole antenna in lossy
media,” International Journal of Antennas and Propagation, vol. 2008, Article ID
5580782, 2008.
105.
E. C. Fear, et al., “Confocal microwave imaging for breast cancer detection:
localization of tumors in three dimensions,” IEEE Transactions on Biomedical
Engineering, vol. 49, pp. 812-22, 2002.
106.
G. Bindu, et al., “Wideband bow-tie antenna with coplanar stripline feed,”
Microwave and Optical Technology Letters, vol. 46, pp. 3, 2004.
107.
P. J. Gibson, “The Vivaldi aerial,” 9th European Microwave Conference,
Brighton, pp.101-105, 1979.
108.
J. D. Langley, P. S. Hall, and P. Newham, “Balanced antipodal Vivaldi antenna
for wideband phase arrays,” IEEE Proceeding Microwave Antenna Propagation, vol.
143, 1996.
186
109.
A. M. Abbosh, H. K. Kan, and M. E. Bialkowski, “Design of compact ultra-
wideband antipodal antenna,” Microwave and Optical Technology Letters, vol. 48,
2006.
110.
K. L. Watson, and V. C. Sundberg, “Broadband ridged horn design,” Microwave
Journal, pp. 96-101, 1964.
111.
B. M. Natras, C. D. McCarrick, and D. P. Kasilingam, “Two numerical
techniques for analysis of pyramidal horn antennas with continuous metallic ridges,”
Proceedings of IEEE International Symposium for Antenna Propagation, vol. 2.
Pp.560-563, 2001.
112.
E. T. Rosenbury, et al., “Low cost compatible wideband antenna,” Editor, U.S.
2002.
113.
R. M. Sega, “Infrared detection of microwave induced surface currents on flat
plates,” Final Technical Report Air Force Academy, 1982.
114.
C. A. Balanis, Antenna Theory Analysis and Design, John Wiley & Sons, New
York, NY, USA, 2nd edition, 1997.
115.
S. Y. Semenov, A. E. Bulyshev, AA. Abubakae, et al., “Microwave tomographic
imaging of the high dielectric-contrast objects using different image-reconstruction
approaches,” IEEE Transactions on Microwave Theory and Techniques, vol. 53, pp.
2284-395, 2005.
187
116.
A. Taflove and S. C. Hagness, Computational Electrodynamics: The Finite-
Difference Time-Domain Method, Norwood, Mass, USA, Atech House, 2nd edition,
2000.
117.
K. D. Paulsen, P. M. Meaney, M. J. Moskowitz, and J. R. Sullivan, “A dual mesh
scheme for finite element based reconstruction algorithms,” IEEE Transactions of
Medical Imaging, vol. 14, pp. 504-514, 1995.
118.
D. Li, “Microwave imaging spectroscopy for breast cancer diagnosis,” PhD,
Thayer School of Engineering, Dartmouth College, Hanover, NH, 2003.
119.
Parker Hannifin Corp: http://www.parker.com.
120.
Copley Controls: http://www.copleycontrols.com.
121.
N. R. Epstein, P. M. Meaney, and K. D. Paulsen, “MR-guided conformal
microwave imaging for enhanced inclusion detection within irregularly shaped
volumes,” SPIE Medical Imaging, 2013.
122.
P. M. Meaney, F. Shubitidze, M. W. Fanning, M. Kmiec, N. R. Epstein, and K. D.
Paulsen, “Surface wave multipath signals in near-field microwave imaging,”
International Journal of Biomedical Imaging, vol. 2012, Article ID 697253, 2012.
123.
M. J. Pallone, P. M. Meaney, and K. D. Paulsen, “Surface scanning through a
cylindrical tank of coupling fluid for clinical microwave breast imaging.” Medical
Physics, 2012.
188
124.
D. Li, P. M. Meaney, and K. D. Paulsen, “Conformal microwave imaging for
breast cancer detection,” IEEE Transactions on Microwave Theory and Techniques,
vol. 51, pp. 1179- 1186, 2003.
125.
N. R. Epstein, A. Golnabi, P. M. Meaney, K. D. Paulsen, “Conformal microwave
tomography using a broadband non-contacting monopole antenna array,” IEEE
Microwave Theory and Techniques Ultra-Wideband Society, 2012.
189
Документ
Категория
Без категории
Просмотров
0
Размер файла
11 946 Кб
Теги
sdewsdweddes
1/--страниц
Пожаловаться на содержимое документа