close

Вход

Забыли?

вход по аккаунту

?

Toward computerized morphometric facilitiesA review of 58 software packages for computer-aided three-dimensional reconstruction quantification and picture generation from parallel serial sections.

код для вставкиСкачать
THE ANATOMICAL RECORD 216:449-470 (1986)
Toward Computerized Morphometric Facilities: A
Review of 58 Software Packages for ComputerAided Three-dimensional Reconstruction,
Quantification, and Picture Generation From
Parallel Serial Sections
D.P. HUIJSMANS, W.H. LAMERS, J.A. LOS, AND J. STRACKEE
Departments of Medical Physics @.P.H., J.S.) and Embryology CW.H.L., J.A.L.), University of
Amsterdam, Academic Medical Center, 1105 AZ Amsterdam
ABSTRACT
This review gives a n inventory of 58 computer-aided three-dimensional reconstruction applications in the domain of biomedical research. It is devoted
to the formulation of a set of recommendations thought to be necessary for improved
performance of software packages in this field. These recommendations can be used
to select packages and to guide future developments of existing reconstruction
systems.
The survey is restricted to three-dimensional reconstructions based upon a series
of parallel sections of a n object. Subjects treated are programming languages, resolution and sampling, input preparation, realignment, local deformation of slices,
numerical quantifications, topological complexity, internal representation, display
complexity (hidden surfaces, shading, smoothing), structure extraction, descriptive
elements, database, data compression, time efficiency of systems and algorithms,
hardware configuration, input devices, input media, interactive aids, display devices, and output devices. Information for this survey comes from articles that
appeared between 1965 and 1985.
APPLICATION AREAS
Serial section reconstruction packages, a s shown in
the listing on page 451, have been developed for use
with objects ranging in size from cubic metres (macroscopic) to cubic micrometers (microscopic). Although
three-dimensional reconstructions (3D-reconstructions)
can also be derived from a tilted series of cross-sections
or projections, this approach is beyond the scope of this
review. Macroscopic applications that have been developed in scanner-based diagnostics and anatomy will
gradually evolve into systems directly sampling in twoor three-dimensions using noninvasive imaging techniques (“virtual sectioning”). Two-dimensional examples are computerized axial tomography (Cook et al.,
1980; Glenn et al., 1979; Huang and Ledley, 1975; Pykett et al., 1983; Udupa, 1981), ultrasound (Brinkley et
al., 1978; 1982a,b; Dekker et al., 19741, and magnetic
resonance imaging (Axel et al., 1983; Kramer et al.,
1981; Lai and Lauterbur, 1981; Mendonca Dias et al.,
1982; Pykett et al., 1983). A direct 3D-sampling example
is the Dynamic Spatial Reconstructor at the Mayo Clinic
(Behrenbeck et al., 1982; Block et al., 1984; Gilbert et
al., 1979; Ritman et al., 1978, 1980; Robb et al., 1974,
1980, 1982; Sinak et al., 1985; Wood, 1979). Macroscopic
structures cannot always be imaged noninvasively owing to the lack of a differential or selective imaging
technique. Furthermore, every enlargement is re0 1986 ALAN R. LISS, INC.
stricted to a two-dimensional (focal) plane, because the
magnification process uses optical or electromagnetic
lenses. In such cases the serial section approach is unavoidable (“real sectioning”). An interesting new development toward the application of higher resolution
techniques is the use of the scanning laser microscope
(Van der Voort et al., 1985). In addition, this application
allows the use of thick sections. With the continually
reducing sampling grid size of magnetic resonance imaging (about 1 mm at present), ever smaller anatomical
objects will fall within the reach of this technique. Yet,
light microscopic and electron microscopic sampling distances are still several orders of magnitude away from
what is technically feasible with the technique today.
Many objects will therefore continue to rely upon the
serial sectioning approach as the only means of creating
a n input series for a n enlarged spatial reconstruction.
VISUALIZATION
The presentation of information contained in a spatial
structure reconstructed from parallel cross-sections has
Received February 2, 1986; accepted June 6, 1986.
D.P. Huijsmans’ present address is Department of Applied Mathematics and Computer Science, University of Leiden, Wassenaarseweg
80, 2300 RA Leiden, The Netherlands.
450
D.P. HUIJSMANS, W.H. LAMERS, J.A. LOS, AND J. STRACKEE
led to the use of illustrations and palpable volumetric
models long before the advent of computers. They are
thought to be indispensible aides in perceiving or feeling
spatial structures. The time-consuming nature of the
illustrator’s or modeller’s work toward a n agreeable result severely hampers the number of illustrations or
models produced.
One of the reasons why computers are being exploited
in this field of science is the ease with which they can
be programmed to manipulate large amounts of spatial
data in order to obtain perspective views from arbitrary
viewpoints and to aid in the manufacturing of tangible
models (Reumann et al., 1985).
The improvement in productivity gained by using a
computer-aided reconstruction and display system compares favourably with that of a n extra illustrator. The
most productive unit would probably be a n illustrator
or modeller using the computer to evaluate the effect on
the display of a particular viewpoint and a specific structural or spatial selection chosen. The use of depth-illusion cues such as the kinetic-depth effect obtained from
animation sequences of a rotating object can be used
routinely. The computer enables one to use visualization
aids whose production would be prohibitive when done
manually.
THE THIRD DIMENSION
In general, one hardly realizes the extra amount of
work that is introduced by the third dimension (the
number of sectionsfslices in a series used for a reconstruction). Procedures that seem to work fast enough
when dealing with only one of the slices should still be
fast when they have to deal with 100 sections or more.
Take the simplest case of displaying contour data extracted from 100 cross-sections; if it takes 1 second to
retrieve and display the contour-points from one of the
slices, the total waiting time would amount to 1.5 minutes. When coordinate transformations are added between retrieval and display, these times have to be
multiplied by about 3 which means 3 seconds per slice
and 5 minutes for the complete pile. When hidden line
elimination is included as well, these times become another 3 times longer, giving waiting times of 9 seconds
per slice and 15 minutes for a hidden-line perspective
view of the whole pile of contours. These times are
realistic times on a minicomputer; on a micro- or personal computer these times again have to be multiplied
by as much as 10, giving 1.5 minutes per slice and 2.5
hours per pile! A 100-fold increase in information gathering, manipulation, and display time seems realistic
when dealing with three-dimensional objects as compared to two-dimensional ones. This not only asks for a
speed-up of calculations but of the velocity of in- and
output data streams as well.
Whereas mainframe users would feel comfortable
about the calculation speed of their computer, this would
in general not be the case for their data transfer velocity
during disk access and picture display. Data transfer
speed is as essential as calculation speed in computeraided three-dimensional reconstruction since insight into
large amounts of complex three-dimensional data can
only be grasped from a picture or from a sequence of
pictures.
EARLIER REVIEWS
A number of review articles on three-dimensional reconstruction have appeared since 1972 (Baxter et al.,
1982; Bloch and Udupa, 1983; Herman, 198333;Katz and
Levinthal, 1972; Levinthal et al., 1974; Macagno et al.,
1979; Ware and LoPresti, 1975). However, to our knowledge no comparative analysis of existing software packages that deal with computerized three-dimensional
reconstruction, image generation, and numerical analysis of serial sections has been published. Especially the
boom of microcomputers gives rise to the emergence of
newly developed, simpler versions of existing approaches. Our view as to the merits and demerits of such
small systems is discussed, considering the scope of the
problems in the different application fields and the time
efficiency of the computer-aided solution offered.
AIM OF THE PRESENT REVIEW
Since this review is intended to guide system selection
of specialists in the fields considered (anatomy, embryology, neuroanatomy, X-ray scanning, ultrasound scanning, and magnetic resonance imaging), user characteristics prevail. Different input-, reconstruction- and
display-modes are defined and ranked in order of complexity. Clearly, the more complex and time efficient the
requirements for a particular application are, the more
sophisticated and costly a computer-aided solution will
be.
The 58 software packages for computer-aided threedimensional reconstruction from serial sections in this
review are presented on the following page. The verticle
headings listing authors’ names and respective numbers
in each table of this paper correspond to specific packages in this listing.
When packages are grouped in order of applications in
neuroanatomy, anatomylembryology and scanner-based
diagnostics, answers to many evaluation criteria look
alike and show a tendency from simple to complex (Tables 1and 2).
The main order of presentation of packages against
criteria in all the tables that follow is therefore application (Apl = neuroanatomy, anatomy, embryol, scanner).
Secondary keys are year of first publication (Ye) and
name of first author (Alph). In the Tables, only those
packages that contain information about the reviewed
subject are mentioned.
Evaluation criteria (about 175 in total) are subdivided
into coherent groups (37 headings). Table 2 gives a n
overview of packages against subject headings: a ctY’’
(yes) in the table indicates that data for the criteria
under that heading could be obtained from the publication(s); a “?” (question mark), indicates that data were
not explicitly mentioned but could be inferred from the
articles, and finally a n empty box stands for data not
mentioned. In addition, in Tables 2 and 4 numbers are
used; their meaning is explained in Table 4.
Software packages are developed in a particular setting. How easily a specific software package can be
adapted for use in your laboratory depends upon a number of circumstances. The amount of specific hardware
needed, the internal computer representation chosen
(point cloud, stick figure, contour pile, surface patches,
or volume elements), and the specific problem dealt with
largely determine whether new problems in other areas
451
TOWARDS COMPUTERIZED MORPHOMETRIC FACILITIES
THE 58 RECONSTRUCTION PACKAGES REVIEWED
first author
year
departmentiinstitute
city
~~
1.Afshar F. et al.
2.Batnitzky S. et al.
3.Bioquant
4.Cahan L.D. et al.
5.Capowski J.J.
6.Chawla S.D. et al.
7.Dorup J . et al.
8.Dursteler M.R. et al.
9.Ellis T.J. et al.
10.Fram E.K. et al.
1l.Fujii S. et al.
12.Geiser E.A. et al.
13.Giorgi C. et al.
14.Glaser E.M. et al.
15. Glaser E.M.
16.Graf von Keyserlingk
- D.
l7.Gras H.
l8.Greenleaf J.F.
19.Herman G.T. et a1
20.Hibbard L. et al.
21.Huijsmans D.P. e t al.
22.Huijsmans D.P. et al.
23.Johnson E.M. et al.
24.Katz L. et al.
25.KONTRON
26.Kriete A. et al.
27.Ledley R.S. et al.
28.Levinthal C. et al.
29.Llinas R. et al.
30.Macagno E.R. et a1
3l.Marino T.A. et al.
32.Matsumoto M. et aI.
33.McIntosh J.R. et al.
34.Moens P.B. et al.
35.Nixon J.V. et al.
36.Overdijk J. et al.
37.Paldino A.M.
38.Palmer R. et al.
39.Perkins W.J. et al.
40.Perkins W.J. et al.
41.Prothero J . et al.
42.Prothero J . et al.
43.Rakic P. et al.
44.Schierenberg E. e t al.
45.Shantz M.J. et al.
46.Sobel I. et al.
47.Stein A. e t al.
48.Stevens J.K.
49.Street C.H. et al.
50.Sullivan P.G. et al.
5l.Sundsten J.W. et al.
52.Thompson R.P. et al.
53.Vannier M.W. et al.
54.Veen A. et al.
55.Wann D.F. et al.
56.Willey T.J. et al.
57.Yelnik J. e t al.
58.Yokoi S. et al.
1982
1981
1984
1975
1977
1981
1983
1977
1980
1982
1980
1982
1983
1965
1979
1983
1983
1982
1977
1984
1983
1984
1983
1972
1982
1984
1983
1972
1975
1979
1980
1977
1979
1981
1983
1978
1979
1982
1979
1982
1982
1974
1974
1984
1978
1980
1984
1980
1983
1983
1983
1983
1983
1977
1973
1973
1981
1983
neurosurg./dental anatomy
diagn. radiology
commercial package
surg. neurology
physiology
physiology/electrical eng.
cell biology
anatomy
biophys.bioenglneuro1ogy
radiology
systems eng./radiology
cardiology
neurosurgheuroradiology
physiology
physiology/anatomy
anatomylneurology
zoology/cell biology
physiologyhiophysics
radiology
physiologyhiochemistry
med.phy sics/anatomy
med physidanatomy
physiology
biology
Zeiss
virology/immunology/anatomy
phy siologyhiophysics
biology
neurobiol./physiol.hiophys.
biology
anatomy/radiology
cardiology
molec.and developm.biology
biology
health science
brain research
neuroscience
radiology
computer science
computer science
biologic structure
biologic structure
neuropath./physiol./neurosc.
molecular biology
technology
biology
animal biology
anatomy/physiology
anatomy
dental anatomy
biologic structure
anatomylbiometry
radiology
biology/developm.biol.
electrical eng./anatomy
anatomy/physiology
neurophy siology
information science
London
Kansas City
Nashville
Bethesda
Chapel Hill
Montreal
Aarhus
Lausanne
London
Durham
Kobe
Florida
Milan
Baltimore
Baltimore
Aachen
Gotti ngen
Rochester
Philadelphia
Hershey
Amsterdam
Amsterdam
Chapel Hill
New York
W.Germany
Wurzburg
Washington
New York
Iowa City
New York
Kansas City
Osaka
Boulder
Ontario
Dallas
Amsterdam
Bronx NY
Berkeley
London
London
Seattle
Seattle
Harvard
Gottingen
Pasadena
New York
Chapel Hill
Philadelphia
Memphis
London
Seattle
Charleston
St Louis
Philadehhia
St LouisLoma Linda
Paris
Nagoya
For a number of packages, information presented is based upon several articles: No. 2 upon Batnitzky
et al., 1981, 1982 and Cook et al., 1980; No. 5 upon Capowski, 1977 and Capowski et al., 1981; No. 17
upon Gras et al., 1983 and Gras, 1984; No. 19 upon Herman et al., 1977,1979,1980;No. 20 upon Kontron,
1982 and Medina et al., 1983; No. 28 upon Levinthal et al., 1972 and Nierzwicki-Bauer et al., 1983a,b;
No. 32 upon Matsumoto et al., 1977, 1981; No. 34 upon Moens et al., 1981 and Dupuy-Coin et al., 1982;
No. 41 upon Prothero et al., 1982, 1985; No. 53 upon Vannier et al., 1983a,b, 1984, 1985, and Trotty et
al., 1985; No. 55 upon Wann et al., 1973 and Cowan et al., 1975; No. 58 upon Yokoi et al., 1983 and
Yasuda et al., 1984.
452
D.P. HUIJSMANS, W.H. LAMERS, J.A. LOS, AND J. STRACKEE
IL
R
5 1 3 4
3
2 4 4
2 2 5 5 5
s
0 Y
F
2 4 / 1 3 1
1 1 1 3 1 1 2 3 5 5
S B 1
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y Y
? Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
Y
Y
Y
Y
Y
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y Y
Y Y
Y
Y
Y
Y Y
Y
Y
Y
Y
Y
Y
Y Y Y Y Y Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
? Y Y Y Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y Y ? Y Y Y
?
Y Y ?
?
? Y ?
Y
Y Y Y
Y
Y
Y
Y
Y
Y
Y
? Y Y
? Y Y
?
Y
?
Y
Y
Y
Y
Y
Y Y Y Y Y
Y
Y
Y
Y
Y
Y
Y
Y
?
?
?
?
Y
Y
Y
Y
Y Y Y
Y
Y
Y
? Y Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
?
Y
Y Y ? ? Y ?
?
? ? Y ?
Y
Y
Y Y Y
Y Y ? Y Y Y Y Y Y Y Y
Y
?
Y
Y
? Y Y ? ? ? Y Y Y Y
Y
Y
Y
Y
Y
Y
Y Y Y Y Y Y Y Y Y
Y
Y
Y
Y
?
Y
Y
Y Y Y ?
Y Y Y Y Y
Y
Y
Y
Y
Y
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y Y
Y
Y
Y Y Y Y Y
Y ? Y Y
Y Y Y
Y
Y
Y Y Y
Y
Y Y Y Y Y
Y Y Y Y Y
Y Y Y Y Y
Y Y Y Y Y
Y ? ? Y
Y
Y Y Y Y Y
Y
Y Y Y Y Y
Y Y Y Y Y
Y ? Y
Y Y Y
Y
Y
Y Y Y
Y
Y
Y
Y Y ?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
?
Y
Y
Y
?
4 ? ? 1 5 ? 5 4 6 ? 6 2 ? 6 4 ? 2 2 4
? 3
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y Y
Y
Y
? ? ? Y ? Y ? ? Y ? Y ? Y Y Y
Y Y Y Y Y Y Y Y Y
Y Y Y Y Y
Y
Y
Y Y Y Y ? Y Y ? Y Y Y Y Y
Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
?
Y
Y Y
Y Y Y Y Y Y ? Y
Y Y
Y
Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
Y
Y
Y
Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
Y ? ? Y ? ?
Y ? ? Y Y ? Y Y
Y ? ? Y ? ?
Y ? ? Y ? Y Y Y
Y Y Y Y Y Y
Y Y Y Y Y ? Y Y
Y
Y
Y Y
Y
Y
Y Y Y
Y Y Y Y Y
Y Y
Y
Y
Y Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
?
Y
Y
Y
Y ? ? Y ? ? ? Y ? Y Y Y ? ? ? Y Y Y
Y Y Y Y Y
Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
7 ? 4 5 5 1 2 2 5 ? ? ? 2 ? ? ? ? 4 ? 5 7 4
Y Y
Y
Y Y
Y Y
Y
Y ?
Y
Y
Y Y
Y
Y
?
?
Y
Y
Y
?
Y
Y
Y
Y Y Y
Y
Y
Y
Y ?
Y
Y Y Y
Y ? Y
Y
Y
Y
Y
Y
Y
Y
?
?
Y
Y
Y Y
Y
Y
Y
Y
?
?
Y
Y
Y
?
?
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
? ?
Y
Y
Y
Y
Y
?
?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
? ? ? ?
Y Y
Y
Y Y
Y Y
Y Y
Y Y
Y
Y
Y
Y
Y
Y
Y Y Y
Y
Y
Y
?
Y
Y Y
Y
Y ?
Y
Y Y Y
Y ?
Y
Y
?
?
Y
Y
Y
Y
Y Y
?
Y Y
Y Y
Y Y
Y Y
.-------------_______________
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
?
Y
Y Y Y Y Y Y
? ?
Y
Y Y Y Y Y Y
Y Y Y Y Y Y
Y Y Y Y Y Y
Y Y Y Y Y Y
Y Y Y
Y ? Y
Y
Y
?
Y Y
? Y ? ? ? ? ? ? ? Y ? ? ? ?
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
5 1 4 2 2 4 1 2 4
COMPUTER SYSTEM
PROCESSOR(S)
INPUT DEVICE(S)
INPUT MEDIUM
TEMPORARY STORAGE
PERMANENT STORAGE
DISPLAY DEVICE(S)
INTERACTIVE DEVICE(S)
OUTPUT DEVICE(S)
LOOKUP TABLES
RESOLUTION MINIMUM OF X,Y,Z
MAINFRAME INTERFACE
LANGUAGE
MICROSCOPIC PREPARATIONS
MACROSCOPIC PREPARATIONS
ALIGNMENT
DIGITIZATION
STRUCTURE EXTRACTION
STRUCTURAL DESCRIPTION
DATA ELEMENT
DATA-COMPRESSION
CORRECTION HISTOTECHN.DISTORT.
MICROSCOPIC APPLICATION
MACROSCOPIC APPLICATION
USE
RECONSTRUCTION TIME
PICTURE GENERATION TIME
DIMENSIONAL DISPLAY CAPABILITIES
TOPOLOGY OF HIDDEN L/S VIEWS
TRANSFORMATIONS
HIDDEN LINES/SURFACE REMOVAL
SHADING
SMOOTHING
STEREO PAIR
ARBITRARY NEW CROSS-SECTION
QUANTIFICATIONS
STATISTICS
4
7 1 8 2 8 4 5 1 6 6 4 1 5 0 1 7 1 3 0 1 2 3 2 4 ~ 9 2 1 2 0 2 8 8 3 6 7 5 3 8
2 3 4 3 3 3 3
4 4 5 6 3 4 9 6 5 0 3 7 9 9 8 5 7 7 9 0 6
..............................................................................
1 2 5 5 4
454
D.P. HUIJSMANS, W.H. LAMERS, J.A. LOS, AND J. STRACKEE
case to speed up the much slower Basic interpreter,
which is standard on many micros.
can be tackled with a n existing software package. A
generally applicable software package for three-dimensional reconstruction should therefore support several
types of representation with the possibility of converting
one type
another*
input favours a contourline representation, graphics
a surface representation, and numeric
a
representation.
Resolution and Sampling
The way computer-aided reconstructions
are built up
amountsto a sampling of the spatial structurewith a
three-dimensional coordinate @id. From a theoretical
point of view, the sampling density should be identical
in all three spatial directions, and the sampling distance
should be based upon the smallest details one chooses to
analyse. The sampling distance should be at least twice
as small as the size of the smallest detail of interest
(sampling theorem of Shannon, 1949). When, for example, a microscope is needed, the object has to be sectioned this sectioning fixes the sampling density in the
third dimension (the z direction). The sampling distance
is equal to the slice thickness. Within the slice the
sampling of the x,y coordinates is determined by the
unit size of the digitizing grid and the size of the enlarged slice. Spatial anisotropies introduced by the particular orientation and unit size of the three-dimensional
sampling grid should be avoided. Once sampled, entered, and realigned, the reconstruction of the object
ideally should not have preserved any artifact that is
due to the arbitrary orientation and unit sizes of the
particular grid used during the input phase. The reso-
METHODS AND SOFTWARE
Programming Languages
Surprisingly little is said in most publications about
the programming language(s) used. Only 38%(22 out of
58) mention their programming language. Fortran and
Assembler are the preferred ones. Assembler (a machine
language) is often used to speed up time-critical, frequently used Fortran modules. None of the packages
mentions Algol or Forth. Only one package is in C or
Pascal, a situation that is bound to change in the foreseeable future. C is a programming language a t the
operating system level between high-level languages
and machine languages. It is often encountered in combination with Unix, a multiuser, multitask operating
system. Pascal is very popular within educational circles. Basic is used on several microcomputer systems
(see Table 3). A Basic compiler is recommended in that
W
A
N
N
C
A
H
A
N
M
A
C
A
G
N
P
A
L
D
I
N
P
E
R
K
I
N
E
L
L
I
S
G
R
A
S
S
T
R
E
E
T
H
U
I
J
S
M
A
N
M
A
T
S
U
M
S
0 .
5
3 3 3
1 4 2 1 5 3 3 2 4
2 2 5
2
5 4 0 7 9 9 7 9 0 1 4 1 4 5 1 7 1 3 1 3 2
3 1
0 0 s
H
I
B
B
A
R
l
i
i
l
I
I
V
E
E
N
M
A
R
I
N
M
O
E
N
S
0
D 1
K
O
N
T
R
O
N
P
R
O
T
H
E
R
D
O
R
U
P
I
J
O
H
N
S
O
N
N
0
I
H
U
I
J
S
M
A
S
U
N
D
S
T
E
N
B
I
O
Q
U
A
N
T
S
G
R
A
F
V
0
T K
________________________________________------------_ _2 _ 6_
x x x
fortran:
algol:
pascal:
basic:
assembler: X
x x
x x
x x
X
x x x
X
x x
X
x
X
X
X
X
S
I
1 5 5 4
x
X
X
C:
other:
x x
x
2 3 4 3 3 3 1 2 2 4 1 4 5 4
2 4
2 2 5 5
.
S B i
0
Y
F
2 4 1 1 3 1
1 1 1 3 1 2 3 5 5
4 5 6 3 4 9 6 5 0 7 9 7 0 6 7 ~ 2 4 6 6 1 5 0 7 1 3 0 1 3 2 4 , 9 2 1 2 0 2 8 8 6 7 5 3 8
TOWARDS COMPUTERIZED MORPHOMETRIC FACILITIES
lution within the slice can be brought into correspondence with the slicing resolution by scaling down the x
and y measurements with a n appropriate factor.
The first row of Table 4 gives information on the
minimum resolution used (in percentage of the size of
the object) in any of the three spatial directions; the
resolution is often much lower in the z direction than it
is in the x and y directions. Illustrations of reconstructed
objects made from about 10 sections digitized on a 1,024
x 1,024 tablet are typical examples found in the literature. An unbalanced sampling will always show up in
the display of a reconstruction since details in the undersampled direction are coarser. It may then divert the
viewer’s attention from the real three-dimensional characteristics of the object’s shape.
Rows 2-4 of Table 4 characterize the resolution of the
tablet or image memory buffers when present. Row 2
gives the horizontal number of grid points or pixels
available; row 3 gives the vertical amount, and row 4
gives the number of frame buffers in each system. In the
last row the number of bits per pixel is given; this
determines the number of grey levels that can be represented a t each point of the digitized image. An eight-bitdeep frame buffer allows for the division of the intensity
into 256 levels between black and white.
Input Preparation Procedures
Reconstructions from serial sections of microscopically
small objects undergo a number of input procedures that
are tabulated in Table 5. The majority of published
procedures do not construct realignment reference marks
before starting to slice the object. In our opinion, the
slicing of a spatial structure in order to reveal and
enlarge its inner structure must be accompanied by the
inclusion of reference marks for realignment afterward.
Otherwise the implicit but broken coherence between
sections can only be restored on a visual (subjective)
basis. Systematic errors can be the result.
A majority of the packages reviewed use photographs
or even transparencies obtained from photographs as
input, instead of directly using the enlarged image of
the mounted serial section. Alternative input routes
avoiding the use of the unnecessary and degrading intermediates are given in the section on “Hardware
Aspects.”
Realignment of Serial Sections
Table 6 lists the realignment procedures. About half
of the packages have no computerized realignment built
in. About one-third of the packages did use alignment
reference marks for manually aligning subsequent sections on the tablet or on the screen. A small minority
uses computerized realignment of serial sections, though
not always through the use of reference marks introduced before slicing has taken place.
In neuroanatomy, reconstruction is usually restricted
to data obtained from various focal plane depths from
only one thick slice. This eliminates the need for realignment of data from a series of sections, but neuronal
network reconstructions will thus often be incomplete.
455
The relative positions of hole marks (obtained by drilling perpendicular to the slicing plane) or line references
(obtained from the rectangular side(s) of the embedding
block) can be used to translate x,y origins from different
slices to a common x,y origin and then rotate the sections along a common z-axis perpendicular to the common x,y origin to align the secondary markeds). This
realignment can be followed by either global scaling or
separate x- and y-scaling obtained from distances between reference markers.
Even more advantagous is the selection of reference
points within the object to be sliced as outside reference
points or planes are frequently relatively large and far
away compared to the size of the structures within the
object to be sliced. This increases the accuracy of the
realignment procedure, especially if differential deformation occurs (see next section).
Corrections for Histotechnically Induced Deformations
Even well aligned reconstructions from serial sections
tend to be topologically disturbed owing to deformations
introduced by the slicing and mounting procedures.
Whereas these histotechnically induced errors usually
remain obscured within the section itself (errors between neighbouring points on surface contours are
strongly correlated), they clearly show up in new crosssections that can be generated by computer in planes
different from the original slicing plane. Errors in position between neighbouring points on a n intersection
plane different from the original one are uncorrelated,
and it is in these planes of intersection that we can get
a n “undisturbed view” of the deformations induced by
the slicing and mounting procedures. Few studies (see
Table 7) even mention this source of error, though it
may render a reconstruction useless owing to the disturbed topology, especially when thin-walled structures
are studied. An example of such a disturbed topology is
given in Figure 1.
Although staining is frequently needed to deliniate
structures of interest, the amount of staining may influence the size of structures detected (Perkins et al., 1979).
Variation in the thickness of slices owing to microtome
slicing will affect optical densities and will subsequently
lead to variation in intersection areas unless properly
corrected. For example, we have found that compression
that is due to sectioning results in a n increase of section
thickness. Especially when numerical analysis is one of
the principal aims of the reconstruction approach, every
conceivable action has to be undertaken to either avoid
or correct these deformation errors. Procedures must be
set up to bring their distorting effect down to sampling
resolution size.
An approach used for light-microscopic studies (Laan
et al., in preparation) is the recording of top views of the
embedding block during the microtome slicing of subsequent sections (the so-called episcopic serial section series) to be used as a reference series in conjunction with
the mounted and stained sections (the so-called diascopic
serial section series). From the deformed diascopic image and the undisturbed reference episcopic image, corresponding point pairs are obtained that define a
D
I
N
O
4
5 1 4
K I V O N S E
I S E W I
E
N
N S K
T
S
S K
I
2 3 4 3 3 3 3
n
N
T
O
S
MACROSCOPIC PREPARATIONS
scanning:
photography:
X
s
Y
X
X
S B
1
X
xxxxxxxxxxxxxx
Y
1 2 3 5 5
6 7 5 3 8
F L O N O
E N 1 1
V Y
E
R
K
Y Y Y Y Y Y Y Y Y Y Y Y
1 5 5 4
2 3 4 3 3 3 4
5 1 4 2 2 4 1 2
5 1 3 4
2 4
2 5 5 5 2 4 1 1 1 1 3
4 5 6 3 4 9 6 5 0 3 7 8 5 7 7 9 0 6 7 ~ 8 8 4 5 1 6 6 1 5 0 7 1 0 1 2 2 4 ~ 2 3 6 5
xx
Y Y
X
4 4 5 6 3 4 9 6 5 0 3 7 9 9 8 5 7 7 9
...................................................................................................................
MICROSCOPIC PREPARATIONS Y Y Y Y Y
Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y
Y Y Y Y Y Y Y Y Y
Y Y Y Y Y Y Y Y Y
Y Y
embedding :
xxxxx xxxxxxxxxxxxxxxx xxxxxxxxx xxxxxxxxx x
staining:
xxxxx xxxxxxxxxxxxxxxx xxxxxxxxx xxxxxxxxx x
references:
X
x x
x
x xxx xxx x
slicing:
xxxxx xxxxxxxxxxxxxxxx xxxxxxxxx xxxxxxxxx x
photography:
x xx xxx x x x
xxx x
x x xx x xxxxxxxx
xx
1 2 5 5 4
S Z N L I A N R N A
E
E C N A D T G
R
Y
S Y Z N
K
O
B E I ; I T S N S I E W N H T K T U J N L D M Q J I M S I N M S E M R
E N E G
A T N ~ N H T E N L L S A R I ~ P S S I S P U S E A U I I
N M
T
R L R I
R E
; T E E R O
A
R Q N E
M O V T S A M R
0
s
E
D
I ~ R L
N S R
A N A E O N A E
T
K
A
I A O E
0
N
N N N T N N
O
Y
F
S
S B
IL R
1 3 1
1 1 1 3 1
2 2 4 1 2 4
5 1 3 4
3
2 4 4
2 2 5 5 5
2 4
9 2 1 2 0 2 8 8 3
0 6 7 1 8 2 8 4 5 1 6 6 4 1 5 0 1 7 1 3 0 1 2 3 2 4
G K W W R C L O S M M P P E S C Y G S H K S ~ L ~ D V G M S C M A K P P D ~ J S S T H BM ~F SB F G G P G G L N V Y
L A A I A A L V ~ A C A E L T A E R T I R T ; E R ~ E L A ~ ~ ~ F E ~A UE A RR E
~ R~ A ~I R~ E ~I AH O I
A T N L K H I E A C I L R L E P L A R B I E ; V O R E A R B A E S N R O R I ~ L N O O I R
~ T J T A I E L O A D X N K
~
~
TOWARDS COMPUTERIZED MORPHOMETRIC FACILITIES
C
A
H
A
N
M
A
C
A
G
N
O
P
E
R
K
I
N
S
S
T
E
V
E
N
S
C
A
P
O
W
S
K
I
S
T
R
E
E
T
I
I
I
!
i
(
I
I
,
S
O
B
E
L
C
H
A
W
L
A
457
A H ;
F U I
S I I
H J !
A S :
R M :
A l
N I
S I
3 3 4
4 1 4
2 ;
4 0 9 8 5 9 1 6 6 1 2 1
geometric unwarping coordinate transformation that
does away with local geometric distortions.
Numerical Quantifications
The extraction of parameters such as volume, area,
curvature, moments, etc. from a reconstructed structure
is still a weak point of all listed packages. Although
computers are ideally suited to deal with numbers, these
packages have largely been used for mere graphical
representations. For a general introduction to the use of
numeric shape descriptors in microscopic studies, one is
referred to Blum (19731, Bradbury (1979), and Lipkin et
al. (1966). Morphometry, i.e., the mathematical description of size, is the main topic of Prothero et al. (1985),
Serra (19821, and Udupa (1981). On the possible shortcomings of stereology, i.e., the inference of three-dimensional properties on the basis of two-dimensional
measures, see De Hoff (1983).
Using Table 8 in comparison with those listing the
graphical features offered (Tables 9 and lo), one can see
that most of the software packages offer either extended
graphic facilities and hardly any quantifications, or a
range of quantifications and statistics with only simple
graphic features. Both extended graphics and parameter
extraction features should be aimed at when setting up
a computerized morphometric facility.
Apart from interactively obtainable 2D- or 3D-measures such as position, distance, angles, and perimeter,
few noninteractively obtainable measures except the
simplest [centre of mass, volume (Cook et al., 1980;
Matsumoto et al., 1981) and surface area] are encountered. To us this seems to be a domain where computer
support can provide one with numerical parameters such
as rotational and higher order moments (Hu, 1962; Sadjani and Hall, 1980) calculated from the mass distribution of either the volume or the surface. Such parameters
have proven their use in physics (mechanics). Especially
for the development of shape during growth, a time
function of suitably normalised moments may serve as
a reliable indicator of normal or disturbed development.
Moments can be calculated so as to be variant or invariant for the specific coordinate system and orientation
C
Fig. I. Contourline reconstruction of transverse sections of a n embryonic rat heart (A) and a computer-generated frontal section of this
reconstruction showing the effects of deformations caused by the histotechnical procedures used 03 and C ) .
chosen. Variant moments can be used to estimate the
specific orientation of a known form in a specific coordinate system. The extraction of topologic properties (De
Hoff et al., 1972) such a s the Euler number is another
type of descriptor of three-dimensional shape that is
amenable for automation.
W
I
L
L
E
Y
R
A
K
I
C
L
L
I
N
A
S
O
V
E
R
D
Y
K
M
A
C
A
G
N
O
M
C
I
N
T
O
S
H
P
A
L
D
I
N
O
P E S C Y H S
E L T A E I T
R L E P L B E
K I V O N B I
I S E W I A N
N
N S K R
S
S K
D
I
distances:
cross-sectional area:
volume/densities:
surface area:
angles:
centre qf gravity:
velocity:
rotational moments:
3D invariant moments:
diameters:
X
X
X
x
..............................................................................................
xx
xxxxx xxxx
X
X
x xx xx
x x x
x
x
X
xxxx
x xx xx
xx
X
xx
X
x
xxxx xx
X
X
X
x x x
X
xx
xxx
x x x
X
X
X
X
x
x
xx
X
W
A
N
N
x
I L P G M S C M K P P D H J S S T B H S
I E R L A O H O O E R O U O U U H I U C
; V O A R B A E N R O R I H L N O O I H
[ I T S I E W N T K T U J N L D M Q J I
~ N H E N L L S R I H P S S I S P U S E
i T E R O
A
O N E
M O V T S A M R
N S R
A N A E O N A E
I H R
I A 0
0
N
N N N T N N
IL
S
S B
1 5 5 4 2 3 3 3 3 3
4
5 2 4 1 2 4 1 3 4
3 2 4 4
2 2 5 5 5
2 4
4 5 6 3 9 6 0 3 7 9 9 8 5 7 0 7 [ 8 2 5 1 6 6 4 5 0 1 7 1 3 0 1 2 3 2 4
G
L
A
S
E
R
X
B
A
T
N
I
T
s
K
Y
F
R
A
M
X
G
E
I
S
E
R
X
X
G
R
E
E
N
L
E
A
F
1
1 1 1
1 2 0 2 8
F
U
J
I
I
xxxxx
X
M
A
T
S
U
M
0
T
O
3
2
x
3
5
N
I
X
O
N
459
W
I
L
L
E
Y
R
A
K
I
C
S
H
A
N
T
Z
S
T
R
E
E
T
K
R
I
E
T
E
5 4 4 4 2
6 3 5 9 6
............................................
HIDDEN LINES/SURFACE REMOVAL
Y Y Y Y Y
distance mask:
X X ? X X
temporal priority:
SHADING
Y
transparant surfaces:
solid surf.(hidden surf.remova1):
X
SMOOTHING
Y
intensity (Gouraud):
normal direction (Phong):
X
Topological Complexity and Dimensionality of
Descriptive Elements
V
E
E
N
M
A
R
I
N
O
A
F
S
H
A
R
K
O
N
T
R
O
N
H
U
I
J
S
M
A
N
J
O
H
N
S
O
N
T
H
O
M
P
S
O
N
H
U
I
J
S
M
A
N
S
C
H
I
E
R
E
N
[
[
[
I
[
I
H
E
R
M
A
N
B
A
T
N
1
T
S
K
Y
F
R
A
M
G P L N V Y
R A E I A O
E L D X N K
E M L O N O
N E E N I I
L R Y
E
I
E
R
I
A
S
S B I
F
5 3
2 2 2 5 2 4 / 1 1 1 3 2 3 5 5
4 1 1 5 1 3 2 2 4 1 9 2 0 8 8 7 5 3 8
------------_------ -____-_-----------Y Y Y Y Y Y Y Y Y
Y Y Y Y Y Y Y Y Y
x 7 1 ?
x ? ? X ?
x
x x x x
x
x
? X
Y
X
X
Y
X
Y
X
Y
X
Y Y
Y Y Y
Y Y
x x
x x x
x x
Y Y
Y
Y Y
x x
X
In diffusion studies, a point cloud is often sufficient.
Only point coordinates and a time value will have to be
recorded. Display complexity is low, no shading or hidden feature removal algorithms are needed. Almost any
display will do. A somewhat costly but attractive display
facility would be a refresh vector display for user-controlled real-time rotations.
In growth studies as well as in diagnostics, a volumetric description is often needed. The simplest descriptor
is a record of the structures in the form of a threedimensional volume occupancy array; the elements are
called voxels (volume elements). Although the volumetric descriptors are mere points (representing the centre
of mass of a unit volume element) the display complexity
is like those for surface patches and is discussed further
on.
Contours are used to represent surfaces or surfaces of
enclosed volumes. The depiction of surfaces and volume
bounding surfaces require the inclusion of hidden line
elimination. Although for simple surfaces a rotating
transparent display (used for wire-frames as well) may
be a sufficient visualization aid, for complex and nested
surfaces the transparent displays fall short.
It is with hidden line/surface/volume calculations that
raster refresh displays rather than vector refresh displays come into focus. Whereas on vector displays individual points can be set (turned on) but alterations
require the whole screen to be cleaned, on raster displays the intensity of each individual point can be
changed independently of all the other within a range
of intensities and colors, often referred to as z-modulation. This fact allows for the implementation of a faster
and simpler hidden feature removal algorithm.
For numeric analysis, contour-line reconstructions are
hardly ever interesting in themselves. They appear on
the input side of approaches that reconstruct three-dimensional surfaces or volumes and should therefore be
considered as a n intermediate stage on the way to a real
surface or volumetric description. They are widely used
since manual input from serial sections is easiest and
most efficient starting from outlines; their widespread
use as a visualization aid is in accordance with the
finding from vision research that three-dimensional information is very effectively grasped from sets of parallel surface contours.
One-dimensional Descriptors
Three-dimensional Descriptors
Although all packages are concerned with three-dimensional reconstructions, the spatial complexity of the
elements used to represent the numerical representation (see Table 9) determined most of the hardware and
software needs. In ascending order of complexity, the
dimensionality of the descriptive elements is a s follows:
zero-dimensional-point, voxel; one-dimensional-node
and arc; two-dimensional-contour, cross-section; threedimensional-surface-patch, polygon-tile.
Zero-dimensional Descriptors
When volumetric/surface characteristics and intenIn neuronal network studies, a graph/tree/stick-figure
sity-shaded
displays with hidden surface removal are
representation of connections (arcs) between neurons and
branch-points (nodes) is usually chosen. Display com- aimed at, one arrives at the most complex software
plexity remains low and is essentially as simple as de- features offered. The intensity-shaded hidden surface
scribed above for point-clouds. The more costly vector display comes closest to the way in which we perceive
refresh display can provide the viewer with excellent spatial structures in real life. The most efficient way to
depth-cues through its user-controlled real-time rotating generate such views is from three-dimensional surface
representations such a s triangular tiles, polygones, or
display of the three-dimensional network.
B-spline patches. The set of boundary voxels from the
volumetric representation may also be taken as a n apTwo-dimensionalDescriptors
propriate surface description.
The search for realistic shaded pictures has given birth
A contour-line representation is a n essentially more
complex graphic entity than stick-figures or point-clouds. to several shading and hidden surface elimination algo-
460
D.P. HUIJSMANS. W.H. LAMERS, J.A. LOS, AND J. STRACKEE
rithms; they are treated in the paragraph on “Complex ation algorithms based upon the boundary voxels of the
volume. The set of boundary voxels that represent the
display characteristics.”
surface can be extracted from the volumetric description
Conversion Between Different Representations
using algorithms published in Arzty et al. (1979, 1981);
A close look at Table 9 shows that most of the software Herman and Liu (1978); Herman et al. (1983a); Liu
packages support only one object description, mostly a (1977); and Udupa (1982). Only a highly restricted set of
contour-line one. However, whereas input procedures normal directions is then used to approximate surface
may favour a contour-line representation, numeric anal- orientations. This results in rather poor quality intenysis may favour a volumetric representation and visual- sity-shaded pictures. There is a need for a n algorithm
isation may favour a surface one. Input, analysis and that converts the voxel description into a surface trianvisual support all have their favoured, i.e., most efficient gular tile description with a greater number of surfacedescriptions, and the representation chosen for each de- normal directions than the six in boundary voxel sets (of
pends upon the input route followed (manual, auto- which only three will be visible from a particular viewmated), the problem a t hand (diffusion, neuronal point). We have not encountered such a voxel-to-surfacenetworks, embryonic growth, changes in shape, diagnos- tile algorithm but are working on it ourselves using the
tics), and the visual support wanted (point-cloud, stick- theory on space fillers, especially those with triangular
figure, wire-frame, contour-line, shaded surface). A gen- faces in order to profit from low-cost specialized harderal-purpose morphometric facility should therefore pro- ware that is (becoming) available.
vide the user with a range of data representations and
At the moment, we calculate approximate normal diconversions between those representations.
rections. A unit normal is assigned to each of the voxel
faces visible. By taking the vectorial sum of 1, 2, or 3
unit normals, a normal direction out of 26 possible direcConversion Between Surface Intersection Contours and
tions
(of which 13 are visible), is obtained. Shading is
Surface Tiles
also used to reflect the depth range by varying the
Since contour representations are frequently met in amount of display intensity from 100% at the front to
the input phase of serial-section reconstructions, and 0% at the back. Range shading on its own is often used
shaded-surface views are aimed at, the first conversions in molecular biology (Van Heel, 1983; Rademacher and
developed were those between a contour-line represen- Frank, 1984) as a simple mean of generating a depth
tation and a triangulated-surface representation. Al- illusion of molecular shapes reconstructed from a series
though in the case of biological shapes a triangulated of tilted projections.
description of the surface involved can always be given,
only one of the published algorithms can deal with more
Complex Display Characteristics
complex topologies. For surface intersections, which
yield a maximum of one intersection contour per sur- The generation of realistic illustrations can include
face, Fuchs et al. (1977); Ganapathy and Dennehy (1982); various forms of hidden lineisurface algorithms, transand Keppel (1973) present triangulation algorithms. parentholid intensity shading, and GouraudPhong inChristiansen and Sederberg (1978) can also deal with tensity smoothing. Table 10 lists characteristics of those
most cases of surface contours near splitting tube junc- packages offering any of these advanced graphic aids.
tions, and Boissonnet (1985) presents a n even more genHidden-feature Removal
eral algorithm that derives a triangulated surface
description from contour sets via the volume enclosed
Hidden-line removal is encountered in closed-contour
by them. With such a triangulated surface description, reconstructions. The area enclosed by the closed-cona system can be configured that makes use of specialized tours in a cross-section is used to suppress or erase
graphic display boards for time-efficient generation of contour-points invisible from a specific point of view.
shaded hidden line displays. Such boards in which rota- The transformed pile of contours remains ordered (in
tions, scaling, translations, and perspective projection either ascending or descending order) along each z direcare included, are developed for Computer Aided Design tion in the oblique view. To build up a hidden-line image
and Computer Aided Manufacture (CAD/CAM).
one updates the intermediate image section by section.
The gradual build-up can proceed from back to front,
Conversion Between Intersection Area Outlines and
using a raster display, by erasing pixels that are within
Intersection AreaNolume
the current cross-sectional area before displaying pixels
If a volumetric description for numeric analysis is on the boundary contours of the current area. This apwanted and input is manual (contour following on a proach is called hidden-line elimination by temporal
tablet), one has to convert the obtained contour-line rep- priority or the painters algorithm. The alternative way
resentation to a volumetric one. In this case the inner- to build up the projected image proceeds from front to
and outer-(closed)contour sets can be converted into the back and uses a second image with the union of crosscorresponding cross-sectional area or volume slice by sectional areas treated so far as a visibility mask, each
algorithms that are described in Huijsmans (1983).
contour pixel of the current section is displayed when
outside the mask, otherwise its display is suppressed
Conversion Between Voxels and Surface Tiles
and finally the mask is updated with the enclosed area
When automated structure extraction from digitized of the current section. This approach is called hiddenimages is used, cross-sectional areas can be extracted line elimination by spatial priority. The spatial priority
directly and used to build up the volumetric description. approach can be used on both vector and raster displays,
To generate shaded hidden surface pictures from such a but the much faster painters algorithm can only be used
voxel representation, one can use simple display gener- on raster displays unless the final image has been con-
1 LR
S
S B 1
0
F
2 5 5 4
2 4 3 3
5 1 4 2 2 1 2
5 1 3
3
2 4
2 2 5
2 4 1 1 3 1 1 1 1 3 1 1 2 3 5 5
4 5 6 3 4 9 5 0 7 5 7 7 9 0 6 1 8 8 4 5 1 6 4 1 5 0 7 1 3 2 3 2 4 1 9 2 1 0 2 8 8 3 6 7 5 3 8
462
D.P. HUIJSMANS, W.H. LAMERS, J.A. LOS, AND J. STRACKEE
structed on disk first before being converted to point
vector commands.
For the generation of hidden-surface representations
the z - b d e r algorithm is the most general solution. It is
used when the sequence of intensities generated are not
sorted on depth. For every generated shading intensity
a t position x,y,z (z being the depth of the display) its zvalue is compared to the z-value of the intensity currently stored a t x,y; if the old z-value is lowest it is
retained, otherwise it is overwritten by the new intensity and z-value (see Griffiths, 1978; and Sutherland et
al., 1974).
For hidden-volume elimination, the z-buffer algorithm
need not be used. The voxel description is already ordered on x, y, and z. Back-to-front (highest to lowest z)
display using the painters algorithm generates the final
image and is especially useful for transparant shapes; a
front-to-back search (lowest to highest z) along each x,y
pixel position for the first occupied voxel and suspending
a further search is the fastest solution for opaque
surfaces.
Shading
Intensity shading calculations can be performed on
opaque as well as transparent objects. In the transparent mode, all the object points along each z-line for pixel
position x,y together determine the shaded intensity a t
pixel x,y. All the object points must then be treated.
Smoothing
Intensity smoothing between neighbouring points on
the surface is used to enhance the quality of generated
images. Two main approaches have been developed: one
according to Gouraud (1971) and one according to Phong
(1975). Gouraud and Phong smoothing both apply to
triangles; in Gouraud‘s case intensit,ies within the triangle are linearly interpolated from intensities at the
corners, whereas Phong calculates normals (direction
vectors perpendicular to the local surfaces) by linear
interpolation of those a t the corners before using a nonlinear angular reflexion formula for the shaded intensity, which is particularly useful when highlighting
reflexions are included. With Phong’s approach one can
produce a n image of normal directions and use the lookup table settings to implement a specific angular reflexion function.
Elementary Coordinate Transformations
Elementary coordinate transformations include any of
the following: rotation, translation, scaling, orthogonal
and perspective projection. Although translation, rotation, and projection are present in most packages, perspective projection is often missing, even in packages
that claim to offer it. To avoid misunderstanding we
tabulated the elementary transformations offered in Table ll.
Structure Extraction and Description
stalled fully automated detection and extraction procedures. All but one use thresholding; in one package
picture subtraction is used as well. A somewhat larger
number uses semiautomated procedures, usually by
user-controlled thresholding (global interaction) and/or
defining contours (local interaction). Most of the studies
cited are still using manual structure extraction procedures. Progress of accepted algorithms from artificial
intelligence into this field is slow. The way representations are described by named elements such as points/
nodes is tabulated as well. A large majority uses some
sort of contour.
Database Elements and Data Compression
In Table 13 a n overview of stored descriptive elements
is given a s well as a n inventory of the memory-saving
techniques used. Descriptive elements such as surface
tiles (polygons/triangles) and volumetric elements (voxels) are scarce. Although spline and other parametric
descriptions for reconstructed objects have been developed, we would like to emphasize that only a limited
class of topological objects can be represented in this
way. For a parametric description, one has to specify a
coordinate transformation that maps the surface onto a
single valued function of the new coordinates everywhere (this class of objects is restricted to the so-called
“star-like” ones; i.e., within the object’s surface a n origin can be found so that every ray going out from the
origin only intersects the surface once, for instance by
adopting a cylindrical or polar coordinate frame with a n
appropriate origin). One should therefore be cautious in
trying to aim a t a n analytic approximation of entered
three-dimensional data of biological shapes. The generation of analytic descriptions for given data sets representing closed surfaces of arbitrary shape still belongs
to the class of mathematically unsolved problems.
Memory-saving techniques are used by the majority of
users. Since the a r c d i n e segments used to define tree
structures and contours are connecting adjacent branchpoints or contour-points, the subsequent nodes of the
connecting arcs are stored, achieving a 50% reduction
in memory use. One of the three spatial coordinate Values, usually the z-value, needs only to be stored once for
each serial section data. A line fit to further lower the
number of chain elements in a contour is used in most
cases. An equal reduction in memory size is achieved by
using a n incremental-addressing technique (chain code)
using 2 (crack code) or 3 (chain code) bits per coordinate
pair instead of 2 words (Freeman, 1961). Both methods
have their advantages and disadvantages. Scaling-entered contours in combination with a line-fit or chaincoding eliminates fine detail that may be due to contourfollowing errors and further reduces the amount of data
space needed. We recommend a line-fit or scaling to
eliminate unnecessary detail. However, when fine details have to be retained, chain-coding without scaling
is advised. Since line-fitting of contours is done independently of those in neighbouring sections, a loose fit can
give the whole reconstruction a bumpy appearance in
the z direction.
Table 12 gives a n overview of the modes of structure
extraction (manual, semiautomated, and automated)
used. It is followed by a list of image processing or
Time Efficiency of Reconstruction and Picture Generation
feature extraction steps employed and the nature of
structural description used in the database. Only a small
It is difficult to assess the time efficiency of reconstrucminority of reconstruction studies claims to have in- tion and picture generation from the papers studied
x
Y
?
?
xxx
x
xx
x
Y Y Y Y
Y Y Y Y Y Y Y
xx
Y Y
Y Y Y Y Y Y ? Y
X
X
xx XX?
?
Y ?
X
Y ? ? Y
X
X
x ?X
X
X
Y Y Y
?
? ? ? ?
1 L
R
s
S B l
0
Y
F
1 2 5 5 4
2 3 4 3 3 3 3
4
5 1 4 2 2 4 1 2 4
5 1 3 4
3
2 4 4
2 2 5 5 5
2 4 1 1 3 1
1 1 1 3 1 1 2 3 5 5
4 4 5 6 3 4 9 6 5 0 3 7 9 9 8 5 7 7 9 0 6 7 ~ 8 2 8 4 5 1 6 6 4 1 5 0 1 7 1 3 0 1 2 3 2 4 ! 9 2 1 2 0 2 8 8 3 6 7 5 3 8
xx
Y Y
?
? ? ?
?
? ?
? ?
? ? ?
X
? Y ? Y Y Y Y
? X
X
? X
x
X
X
x
?
? ?
Y ?
? X
?
Y ? ? Y ? ?
X
? ? X ? ?
?
? ?
xx
X
X ?
Y ?
Y Y ? Y Y Y Y Y
? ?
?
x
X ?
xxxx
? ? ?
xxx
X ? X X
X X X ?
X
X
Y ? Y Y
xx
Y Y ? Y Y Y Y Y Y Y Y ?
Y ? ? Y ? Y Y Y Y
x
Y ?
X
X X ? X X
Y Y Y ? Y Y
X
?
Y Y ? Y ? ? ? ? ? Y Y ? Y Y
X ?
X
X ?
? ? ? X ? X X
x
X
?
?
? ? ? ? ?
X X ? ? ?
?
Y Y ?
x x
...............................................................................................................................................
RECONSTRUCTION TIME
Y Y Y Y ? Y Y Y ?
minutes:
X
hours:
x x
?
days :
?
X ?
x ?
weeks :
PICTURE GENERATION TIME Y Y Y Y Y Y ? Y ?
video rate (1/30 sec):
x
X ?
minutes:
X ?
xx
X ?
hours :
...............................................................................................................................................
patches :
DATA-COMPRESSION
chaincode:
scaling:
line fit:
1 L
R
s
S B I
0
Y
F
1 2 5 5 4
2 3 4 3 3 3 3
4
5 1 4 2 2 4 1 2 4
5 1 3 4
3
2 4 4
2 2 5 5 5
2 4 1 1 3 1
1 1 1 3 1 1 2 3 5 5
4 4 5 6 3 4 9 6 5 0 3 7 9 9 8 5 7 7 9 0 6 7 [ 8 2 8 4 5 1 6 6 4 1 5 0 1 7 1 3 0 1 2 3 2 4 ~ 9 2 1 2 0 2 8 8 3 6 7 5 3 8
464
D.P. HULJSMANS. W.H. LAMERS, J.A. LOS, AND J. STRACKEE
since it is usually not explicitly mentioned. The times
tabulated in Table 14 are therefore coarse estimates that
may vary considerably between reconstructions.
Picture generation times are better known. Although
real-time image generation speed would be welcomed by
all, this speed can a t the moment only be offered for
point-cloud and wire-frame representations on a vector
refresh display. Ordinary vector displays take much
longer, especially when operated in a multi-user environment. The inclusion of hidden-feature elimination or
the use of shading slows down generation times by orders of magnitude.
A stand-alone minicomputer with additional special
purpose video boards and pipelined array processor
boards for image display and coordinate transformations
offering high calculation and transfer speed has a good
pricelperformance ratio.
HARDWARE ASPECTS
Computer System Characteristics
Table 15 starts with a division of computing power
into the categories mainframe, mini-, and microcomputer systems. A trend over the years from mainframe to
mini and micro can be detected in applications in neuroanatomy and anatomy/embryology. User-made hardware used in some of the older applications is not so
important anymore.
Although video boards, image memories, and array
processors for the speed-up of time-critical functions are
attractive features, they are not used by the majority.
A number of permanent storage devices or media-like
punched cards and papertape have become obsolete and
have given way to floppies and hard disks. Especially
hard disks are attractive storage devices for reconstruction databases and pictures.
Since communication between computers is getting
more and more important, information on interfaces is
shown as well. The large data volumes connected with
three-dimensional data and images favour a fast (parallel) interface.
Finally, the main digitizing hardware is inventoried.
The tablet scores high in all categories; scan conversion
is restricted to the macroscopic applications. Low-cost
systems prevail in the field of anatomy (microscopic
applications), and high-cost extensions prevail in scanner-based diagnostics.
Input Devices and Input Media
ceeds. In (semi)automated systems, it usually follows
digitization.
Several circumstances give rise to the use of intermediate storage of the enlarged analog images. In electron
microscope studies, filmstrips are often used; in light
microscope studies, photographs are used. Often a n intermediate is used as a n interface between the researchers and the computer department. When structures are
manually extracted, contour definition and naming is
sometimes done as a separate stage, with photographs
as input and transparencies as output series. There is a
loss of accuracy associated with any intermediate added
along the input route. An alternative way of defining
contours from mounted sections while retaining accuracy is the use of a superposition device such as the
camera lucida, using the combined images of the enlarged section and the tablet as a guide during the
manual contour-following. A further advantage of working with the enlarged image of the mounted section
itself is that illumination conditions can be optimized
during the structure extraction phase. When changing
illumination conditions with filters, diaphragm and
lamp voltage settings are best kept constant, especially
when image subtraction is used.
In video-based digitizations, the camera can be
mounted on top of the microscope, and contours can then
manually be defined by superposing a cursor on the
digitized image to guide mouse or tablet movements.
Here too, there is no need for photographs to store the
enlarged image of the section.
In scanner-based studies, photographs produced by the
firmware often serve as input series for a reconstruction
either because users are unaware of the redigitization
this amounts to or because the manufacturer is not
willing to disclose the memory-saving coding algorithm
for CT-number storage. This of course should be avoided
at any price; direct access to the matrices of the CTnumbers must be provided by the firm(ware).
Interactive Aids
Interactive aids are tabulated in Table 17. They can
be divided into two classes: point and continuous curve
interaction. Pointers can be seen as zero-dimensional
interactive aids and the contourers as one-dimensional
ones. Approaches toward reconstruction, unless entirely
consisting of point-clouds or branch-points, are bound to
need a contouring device either to define contours or to
assist semiautomated contour extraction. Since contourers can also be used as pointers, it is the appropriate
interactive device in reconstruction software. The use of
either a mouse or a tablet sflices. The tablet delivers
absolute coordinates that can be used to detect relative
changes as well; the mouse delivers coordinate changes,
which in combination with a screen superposed cursor
can also do absolute addressing. The third contouring
device, the lightpen, for interaction directly a t the screen
is reported to be tiresome when frequently needed as a
contourer or pointer. A microphone or footswitch is
sometimes used in complex interaction to free the hands
of the operator.
Computer-aided three-dimensional reconstruction
needs some device to transform analog spatial or pictorial information into numerical (digital) form. Usually
a n intermediate analog device must be used to enlarge
the cross-section of the object in order to reach a n acceptable object resolution given the fixed sampling grid a t
hand. The analog-to-digital converter (AD-converter) or
digitizer can be a scanner (CT, X-ray, MRI, or drum), a
video frame digitizer, a tablet, a mouse, a lightpen, a
joystick, or a (fo0t)switch. Scanners and frame digitizers
create a two-dimensional matrix of numbers (intensity
levels). Tablet and mouse can be used to enter continuous curves as well as individual points. Cursor, lightpen,
Picture Display and Output Devices
joystick, and switches are point-entering devices (see
Table 16). Whether digitization preceeds structure exThe devices and output media listed in Table 18 fall
traction differs from one project to another. In manually into two classes: (temporary) display devices and (perdefined reconstructions, structure extraction often pre- manent) storage media.
W
A
N
N
W
I
L
L
E
Y
R
A
K
I
C
C
A
H
A
N
L
L
I
N
A
S
O
V
E
R
D
Y
K
S
H
A
N
T
Z
X
X
?
X
X
?
7
x x x
xx
x
X
X
X
x
X
X
x
x
x
? X
x x
xxxx
x
X
X
X
x
X
xx
x
x
x
x
x
x
X
X
x
X
X
xx
xxx
xx
X
x
X
X
X
x
X
X
X
x
x
xxx
x
x
x
X
x
xx
X
x
xx
X
x
xx
x
x
xxx
X
X
x
X
x
xx
xxx
X
xxxxxxxxxx
X
X
xxx
x
x
X
x
xx
xxx x x
xx x
xxx
x
X
x
X
xxx
x
? X X ?
X
1L
R
S
S B 1
0
Y
F
1 2 5 5 4
2 3 4 3 3 3 3
4
5 1 4 2 2 4 1 2 4
5 1 3 4
3
2 4 4
2 2 5 5 5
2 4 1 1 3 1
1 1 1 3 1 1 2 3 5 5
4 4 5 6 3 4 9 6 5 0 3 7 9 9 8 5 7 7 9 0 6 7 1 8 2 8 4 5 1 6 6 4 1 5 0 1 7 1 3 0 1 2 3 2 4 ~ 9 2 1 2 0 2 8 8 3 6 7 5 3 8
------_-------____-___________
camera lucida:
tablet :
video camera:
drum scanner:
scanner :
microscope:
film camera:
object:
mounted slice:
photograph:
negative/rnicrograph:
transparancy:
filmstrip:
atlas:
video tape:
mainframe :
m:ni:
micro:
special purpose hardware: X
general purpose:
video processor:
array processor:
frame buffer:
punched cards/papertape:
tape:
x
f pPPY:
disk:
serial interface:
parallel interface:
tablet:
AD-conversion:
X
video AD conversion:
.................................................................................................................................................
x xx x
X
X X ?
X
X
?
X
?
? X
x
x xxx xxxxx
xxx x
x xx
x
xx x
xx
xx xxxx
? X
x xxx
x x x
xxx
X
X
X
xxxxxxxxxxxxxxxxxxxxx xxxx xxxx xxxxxxxxxxxx XXX?XXXX xxxxx
X
xx
xx
X
?
x x
xx
x xx
xx
X
X
x x
xx
xxx
X ?
x
x
xx x
? X
x xx
xx
X
xx
X
X
X
x
x
x
x x
X
X
x x xx x
xx
x
xx x x x
xxx
X
X
xx xx xxx
xxx
x
x x
xx xxx
xxx xx
X
?
?
X
x
xxxx
X
X
xxxxxxx x
xxx
xx
xx xx xxxxx?xxxx
xx x
x
x xx x
xxx
X
X
xx
X
K
A
T
Z
M M P P E S C Y G S H K S ~ L P D V G M S C M A K P P D H J S S T B H S [ H M F B F G G P G G L N V Y
A C A E L T A E R T I R T ~ E R U E L A O H O F O E R O U O U U H I U C ~ E A U A R E R A I R E I A O
C I L R L E P L A R B I E ~ V O R E A R B A E S N R O R I H L N O O I H ~ R T J T A I E L O A D X N K
A N D K I V O N S E B E I [ I T S N S I E W N H T K T U J N L D M Q J I [ M S I N M S E M R F L O N O
G T I I S E W I
E A T N l N H T
E N L L S A R I H P S S I S P U S E [ A U I I E N E G
E N 1 1
N O N N
N S K
T R E
/ T E E R O
A
R O N E
M O V T S A M R I N M
T
R L R I V Y
E
O S O S
S K
D
1 H R L
N S R
A N A E O N A E I
0
S
E
R
H
I
i A O E
0
N
N N N T N N !
T
K
A
K
R
S
S B I
0
Y
F
IL
1 2 5 5 4
2 3 4 3 3 3 3
4
5 1 4 2 2 4 1 2 4
5 1 3 4
3
2 4 4
2 2 5 5 5
2 4 1 1 3 1
1 1 1 3 1 1 2 3 5 5
4 4 5 6 3 4 9 6 5 0 3 7 9 9 8 5 7 7 9 0 6 7 ~ 8 2 8 4 5 1 6 6 4 1 5 0 1 7 1 3 0 1 2 3 2 4 [ 9 2 1 2 0 2 8 8 3 6 7 5 3 8
G
L
A
S
E
R
I
I
.................................................................................................................................................
.................................................................................................................................................
ANATOMY/EMBRYOLOGY
SCANNER DIAGNOSTICS
APPLICATION FIELDS:
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .NEUROANATOMY
........................................................................................................
TABLE 15: COMPUTER SYSTEM HARDWARE CHARACTERISTICS OF THREE-DIMENSIONAL RECONSTRUCTION
vector B/W display:
plotter :
printer:
x
x
xx xx x xxx
x xx
xx
xxx xxxx x
? X
xx
x
x
xxx
xx
x
x xxx
?
xx
xxx xx
x
x
? ?
x
xx x
xxxxxxxx xx
x x xx xxxx
xxxxxxx
X
x
xxx x
X
X
xxxx
TOWARDS COMPUTERIZED MORPHOMETRIC FACILITIES
Displays can be divided into vector and raster displays. Vector displays are point- and line-drawing devices especially suited to display point-cloud, stick-figure,
and wire-frame reconstructions that use neither hiddenfeature elimination nor shading. Color is usually absent
on vector displays, although expensive units may provide them.
Raster displays, either monochrome (B/W) or color,
show the contents of a continually refreshed digital image, a matrix of image points (pixels) with a range of
intensities andfor colors for each individual point. They
are especially suited to depict shaded and hidden-surface
representations.
Within a few years, digital optical disks for singleframe recording and real-time playback will probably
become available as useful screen copies.
CONCLUSIONS
From the present study, we conclude that a t present
hardware facilities poorly reflect the requirements of
computer-aided three-dimensional work, because such
work takes a factor hundred times longer than the
equivalent two-dimensional work. Therefore, a standalone computer, with both fast calculation speed and
fast data transport, is to be preferred. For input and
display, special-purpose video boards, image AD/DA converters, lookup tables, image memories, a n image array
processor, and a pipelined array processor for floatingpoint coordinate transformations may have to be included into the system. In addition, a contour and pointing device such a s a tablet is needed for user interaction.
Surface or volume representations with color, shading,
smoothing, and hidden-feature removal is most efficiently done on a raster-refresh display, whereas pointcloud and stick-figure representations are most efficiently done on a vector-refresh display.
From a theoretical point of view, numeric analysis and
display should be based upon reconstructions that aim
a t equalizing the sampling industry in all three directions. Especially the third dimension (the number of
slices) is undersampled in almost all of the reviewed
applications. Realignment reference marks should be
introduced before slicing the object, and the realign and
scaling procedure should be automated. If needed, episcopic reference series can be used as a n aid in correcting
for histotechnically induced global and local deformations. The use of intermediates for the enlarged optical
image should be avoided whenever possible.
A software package with extended graphic facilities
and extended numerical analysis support should be
aimed at. Conversion between different representations
may be needed to satisfy input, output, and analysis
needs. Manual input is easiest with contours; realistic
shaded images favour a triangulated surface description, and numeric analysis favours a volumetric description. Since it is possible to derive all other representations from a volumetric representation it is the
preferred representation. Database names should provide entry on both global (3D) and descriptive element
level (2D/lD). Contours can be stored efficiently using a
line-fit or a crack-code. Smoothing the data set is best
done on the realigned three-dimensional shape taking
all three dimensions into account.
Apart from our findings summarized above, we would
like to point out that none of the packages reviewed is
467
finished (yet). Given the diverse applications and methodologies inventoried we are now in the position to define the elements of a generally applicable software
package for reconstruction, graphics, and numerical
analysis. The three software packages presently being
sold commercially-Kontron Mop Videoplan (Carl Zeiss,
Oberkochen, FRG), Bioquant Image Analysis System
(R & M Biometrics, Nashville, TN), and Eutectic Electronics (Raleigh, NC) are far from ideal. The picture
generation time of the Kontron package, which is developed by Digital Research, is almost prohibitively slow
(especially when dealing with hidden line elimination,
presumably because each intersection area to be displayed is compared to each of the sections in front of it
instead of the union of section areas in front of it).
Although Bioquant offers modules for morphometrics,
its contour-line model is a simple transparent one solely
for display purposes. The graphics of Eutectic Electronics, which is a commercialized version of Capowski and
Sedivic’s package (1981), is better and faster, but what
all commercial packages lack is the ability to convert
contour-line representations into volumetric ones and to
obtain extensive numerical quantifications from either
representation.
ACKNOWLEDGMENTS
This research was made possible by grants from the
Netherlands Heart Foundation and the Netherlands Organisation for the Advancement of Pure Research
(ZWO).
LITERATURE CITED
Afshar, F., and E. Dykes (1982) A three-dimensional reconstruction of
the human brain stem. J. Neurosurg., 57:491-495.
Arzty, E. (1979) Display of three-dimensional information in computer
tomography. Comput. Graph. Image Process., 9.196-198,
Arzty, E., G. Frieder, and G.T. Herman (1981) The theory, design,
implementation and evaluation of a three-dimensional surface detection algorithm. Comput. Graph. Image Process., 15.1-24.
Axel, L., G.T. Herman, J.K. Udupa, P.A. Bottomley, and W.A. Edelstein (1983) Three-dimensional display of nuclear magnetic resonance (NMR) cardiovascular images. J. Comput. Ass. Tom., 7.172174.
Batnitzky, S., H.I. Price, P.N. Cook, L.T. Cook, and S.J. Dwyer I11
(1981) Three-dimensional computer reconstructions from surface
contours for head CT examinations. J. Comput. Ass. Tom., 5.60-67.
Batnitzky, S., H.I. Price, K.R. Lee, P.N. Cook, L.T. Cook, S.L. Fritz,
S.J. Dwyer 111, and C. Watts (1982) Three dimensional computer
reconstructions of brain lesions from surface contours provided by
computed tomography: A prospectus. Neurosurgery, 11:73-84.
Baxter, B., L.E. Hichner, and R.E. Anderson (1982) Application of a
three-dimensional display in diagnostic imaging. J. Comput. Ass.
Tom., 6:lOOO-1005.
Behrenbeck, T., J.H. Kinsey, L.D. Harris, R.A. Robb, and E.L. Ritman
(1982) Three-dimensional spatial, density and temporat resolution
of the Dynamic Spatial Reconstructor. J. Comput. Ass. Tom.,
6.1138-1147,
Bloch, P., and J.K. Udupa (1983) Application of computerized tomography to radiation and surgical planning. Proc. IEEE, 71:351-355.
Block, M., Y. Liu, L.D. Harris, R.A. Robb, and E.L. Ritman (1984)
Quantitative analysis of a vascular tree model with the Dynamic
Spatial Reconstructor. J. Comput. Ass. Tom., 8:380-400.
Blum, H. (1973) Biological shape and visual science (Part I). J. Theor.
Biol., 38:205-287.
Boissonnat, J.D. (1985) Surface reconstruction from planar cross-sections. IEEE Pat. Recogn. Conf. Proc. 7, pp. 393-397.
Bradbury, S. (1979) Micr&copical image analysis: Problems and approaches. J. Microsc., 115.137-150.
Brinkley, J.F., W.D. McCaIlum, S.K. Muramatsu, and D.Y. Liu (1982a)
Fetal weight estimation from ultrasonic three-dimensional head
and trunk reconstructions: Evaluations in vitro. Am. J. Obstet.
Gynecol., 144:715-72 1.
468
D.P. HUIJSMANS, W.H. LAMERS, J.A. LOS, AND J. STRACKEE
Brinkley, J.F., W.E. Moritz, and D.W. Baker (1978) Ultrasonic threedimensional imaging and volume from a series of arbitrary sector
scans. Ultrasound Med. Biol., 4t317-327.
Brinkley, J.F., S.K. Muramatsu, W.D. McCallum, and R.L. Popp (1982b)
In vitro evaluation of a n ultrasonic three-dimensional imaging and
volume system. Ultrasonic Imaging, 4r126-139.
Cahan, L.D., and B.T. Trombka (1975)Computer graphics three-dimensional reconstruction of thalamic anatomy from serial sections.
Comp. Prog. Biomed., 5:91-98.
Capowski, J.J. (1977) Computer-aided reconstructions of neuron trees
from several serial sections. Comp. Biomed. Res., 10:617-629.
Capowski, J.J., and M.J. Sedivic (1981) Accurate computer reconstruction and graphics display of complex neurons utilizing state-of-theart interactive techniques. Comp. Biomed. Res., 14.318-532.
Chawla, S.D., L. Glass, and J.W. Proctor (1981) Three-dimensional
reconstruction of disseminated cancer nodules. Cancer Biochem.
Biophys., 5.153-161.
Christiansen, H.N., and T.W. Sederberg (1978) Conversion of complex
contour line definitions into polygonal element mosaics. ACM8iggraph. Proc. Comp. Graphics, 12:187-192.
Cook, L.T., P.N. Cook, K.A. Lee, S. Batnitzky, B.Y.S. Wong, S.L. Fritz,
J. Ophir, S.J. Dwyer 111, L.A. Bigongiari, and A.W. Templeton
(1980) An algorithm for volume estimations based on polydral
approximation. IEEE Trans. Biomed. Eng., 27r493-500.
Cowan, W.M., T.A. Woolsey, D.F. Wann, and M.L. Dierker (1975) The
Computer Analysis of Golgi-Impregnated Neurons. Golgi Cent.
Svmo. Proc. Santini. ed.. Raven Press. UD. 81-85.
DeHoff, R.T. (1983) Quantitative serial seciioning analysis: Preview. J.
Microsc., 131r250-263.
DeHoff, R.T., E.H. Ailtinger, and K.A. Craig (1972) Experimental determination of the topological properties of three-dimensional microstructures. J. Microsc., 95:69-91.
Dekker, D.L., R.L. Piziali, and E. Dong (1974) A system for ultrasonically imaging the human heart in three dimensions. Comp. Biomed.
Res., 7544-553.
Dorup, J., G.K. Anderson, and A.B. Maunsbach (1983) Electron microscope analysis of tissue components identified and located by computer-assisted 3-D reconstructions: Ultrastructural segmentation
of the developing human proximal tubule. J. Ultrastruct. Res.,
85t82-94.
Dupuy-Coin, A., M. Bouteille, P. Moens, and cJ. Fournier (1982) Threedimensional distribution of nuclear organelles in measles virus
induced polykaryons. Biol. Cell., 4355-68.
Dursteler, M.A., C. Blakemore, and L.J. Garey (1977)Uptake of horseradish peroxidase by geniculocortical axons in the golden hamster:
Analysis by computer reconstruction. Exp. Brain Res., 29:487-500.
Ellis, T.J., D. Rosen, and J.B. Cavanagh (1980) Automated measurements of peripheral nerve fibres in transverse section. J. Biomed.
Eng., 2272-280.
Fram, E.K., J.D. Godwin, and C.E. Putman (1982) Three-dimensional
display of the heart, aorta, lungs and airway using CT. AJR,
139r1171-1176.
Freeman, H. (1961) On the encoding of arbitrary geometric configurations. IRE Trans. Electr. Comp., 10:260-268.
Fuchs, H., Z.M. Kedem, and S.P. Uselton (1977)Optimal surface reconstruction from planar contours. Commun. ACM., 20r693-702.
Fujii, S., Y. Kaneda, M. Matsuo, M. Yoshida, Y. Kobashiri, and H.
Mure (1980) Computer-aided three-dimensional reconstructions and
imaging system for CT images. In: Medinfo 80. D.A.B. Lindberg,
S. Kaihara, eds. North Holland, pp. 185-189.
Ganapathy, S., and T.G. Dennehy (1982)A new general triangulation
method for planar contours. Comput. Graph., 16r69-75.
Geiser, E.A., M. Ariet, D.A. Conetta, S.M. Lupkiewitz, L.G. Christie,
and C. Conti (1982) Dynamic three-dimensional echocardiographic
reconstruction of the intact human left ventricle: Technique and
initial observations in patients. Prog. Cardiol., 103r1056-1065.
Gilbert, B.K., L.M. Kreuger, A. Chu, E.L. Ritman, E.E. Swartlander,
and D.E. Atkins (1979) Applications of optimized parallel processing digital computers and numerical approximation methods to the
ultra high-speed three-dimensional reconstruction of the intact
thorax. Int. J. Biomed. Comput., 1Or317-329.
Giorgi, C., G, Broggi, G. Garibotto, A. Passerini, U. Cerchiari, M.G.
Abele, and M. Koslow (1983) Three-dimensional neuroanatomic
images in CT-guided stereotaxic neurosurgery. ANJR., 4:719-721.
Glaser, E.M., and H. van de Loos (1965) A semi-automatic computermicroscope for the analysis of neuronal morphology. IEEE Trans.
Biomed. Eng., 1222-31.
Glaser, E.M., H. v.d. Loos, and M. Gissler (1979)Tangential orientation
and spatial order in dendrites of cat auditory cortex: A computer
microscope study of Golgi-impregnated material. Exp. Brain Res.,
36r411-431.
Glenn, W.V., M.L. Rhodes, E.M. Altschuler, L.L. Wiltse, C. Kostanek,
and Y.M. Kuo (1979) Multiplanar display computerized body tomography applications in the lumbar spine. Spine, 4.282-352.
Gouraud, H. (1971) Continuous shading of curved surfaces. IEEE Trans.
Comp., 20r623-629.
Graf von Keyserlingk, D., R. de Bleser, and K. Poeck (1983) Stereographic reconstruction of human brain CT series. Acta Anat.
(Basel),115t336-344.
Gras, H. (1984) A hidden-line algorithm for 3D-reconstruction from
serial sections; a n extension of the neurec program package for a
microcomputer. Comp. Prog. Biomed., 18:217-226.
Gras, H., and F. Killman (1983) NEUREC-a program package for 3Dreconstruction from serial sections using a microcomputer. Comp.
Prop.. Biomed.. 17:145-156.
Greenlgaf, J.F. (1982) Three-dimensional imaging in ultrasound. J.
Med. Syst., 6:579-589.
Griffiths, J.G. (1978): Bibliography of hidden-line and hidden-surface
algorithms. Comput. Aided Design, 1Or203-206.
Herman, G.T. (198313) Medical image reconstruction and display. In:
Medinfo 83. v. Bemmel, Ball, and Wigertz, eds. North-Holland, pp.
326-329.
Herman, G.T., and C.G. Coin (1980) The use of three-dimensional
computer display in the study of disk disease. J. Comput. Ass.
Tom., 4t564-567.
Herman, G.T., and H.K. Liu (1977) Display of three-dimensional information in computed tomography. J. Comput. Ass. Tom., 1.155-160.
Herman, G.T., and H.K. Liu (1978) Dynamic boundary surface detection. Comput. Graph. Image Proces., 1:130-138.
Herman, G.T., and H.K. Liu (1979) Three-dimensional display of human organs from computed tomograms. Comp. Graph. Image
Proces., 9: 1-2 1.
Herman, G.T., and D. Webster (1983a) A topological proof of a surface
tracking algorithm. Comput. Graph. Image Proces., 23r162-177.
Hibbard, L.S., and R.A. Hawkins (1984) Three-dimensionalreconstruction of metabolic data from quantitative autoradiography of rat
brain. Am. J. Physiol., 247r412-419.
Hu, M.K. (1962)Visual pattern recognition by moment invariants. IRE
Trans. Inform. Theory, IT-8.179-187.
Huang, H.K., and R.S. Ledley (1975) Three-dimensional image reconstruction from in vivo consecutive transverse axial sections. Comp.
Biol. Med., 5165-170.
Huijsmans, D.P. (1983) Closed 2D contour algorithms for 3D reconstruction. In: Eurographics 83, T. Hagen, ed. North-Holland, pp.
157-168.
Huijsmans, D.P., W.H. Lamers, J.A. Los, J. Smith, and J. Strackee
(1984) Computer-aided three-dimensional reconstruction from serial sections: A software package for reconstruction and selective
image generation for complex topologies. Eurographics 84. Bo
Tucker, eds. North-Holland, pp. 3-13.
Johnson, E.M., and J.J. Capowski (1983)A system for the three-dimensional reconstruction of biological structures. Comp. Biomed. Res.,
16:79-87.
Katz, L., and C. Levinthal (1972) Interactive computer graphics and
representation of complex biological structures. Annu. Rev. Biophys. Bioeng., 1:465-504.
Keppel, E. (1973) Approximating complex surfaces by triangulation of
contour lines. IBM J. Res. Dev., 19t2-11.
Kramer, D.M., J.S. Schneider, A.M. Rudin, and P.C. Lauterbur (1981)
True three-dimensional Nuclear Magnetic Resonance zeugmatographic images of human brain. Neuroradiology, 21:239-244.
Kriete, A., H.J. Wagner, M. Haucke, B. Gerlach, H. Harms, and H.M.
Aus (1984) Dreidimensionale Rekonstruktion elektronen-mikroskopischer Serienschnitten zur Erfassung synaptischer Plasticitat.
Mikroskopie., 41:192-197.
Laan, A.C., D.P. Huijsmans, W.H. Lamers, J.A. Los, and J. Strackee
(1986) A computer-aided input procedure for a volumetric 3-D reconstruction from that restores coherence and removes histotechnically induced geometric distortions (in preparation).
Lai, C.M., and P.C. Lauterbur (1981) True three-dimensional image
reconstruction by nuclear magnetic resonance zeugmatography.
Phvs. Med. Biol.. 26r851-856.
Ledlei R.S., P. Lee, and C.M. Park (1983) High resolution molded
surface view reconstructed from successive CT scans. In: Medinfo
83. T.H. v. Bemmel, M.T. Ball, 0. Wigertz, eds. North-Holland, pp.
353-354.
Levinthal, C., E. Macagno, and C. Tountas (1974) Computer-aided
reconstruction from serial sections. Fed. Proc., 332336-2340,
Levinthal, C., and A. Ware (1972) Three-dimensional reconstruction
from serial sections. Nature, 236r207-210.
Lipkin, L.E., W.C. Watt, and R.A. Kirsch (1966) The analysis, synthesis
and description of biological images. Ann. N.Y. Acad. Sci., 128:9841012.
TOWARDS COMPUTERIZED MORPHOMETRIC FACILITIES
Liu, H.K. (1977) Two- and Three-dimensional boundary detection.
Comput. Graph. Image Proces., 6:123-134.
Llinas, A., and D.E. Hillman (1975) A multipurpose tridimensional
reconstruction computer system for neuroanatomy. Golgi Cent.
Symp. Proc. Santini, ed., Raven Press, New York, pp. 71-79.
Macagno, E.A., C. Levinthal, and I. Sobel (1979) Three-dimensional
computer reconstruction of neurons and neuronal assemblees.
Annu. Rev. Biophys. Bioeng., 8:323-351.
Marina, T.A., P.N. Cook, L.T. Cook, and S.J. Dwyer I11 (1980) The use
of computer imaging techniques to visualize cardiac muscle cells
in three dimensions. Anat. Rec., 198:537-546.
Matsumoto, M., M. Inoue, S. Tamura, K. Tanaka, and H. Abe (1981)
Three-dimensional echocardiography for spatial visualization and
volume calculation of cardiac structures. J. Clin. Ultras, 9t157-165.
Matsumoto, M., H. Matsuo, A. Kitabatake, M. Inoue, Y. Hamanaka, S.
Tamura, K. Tanaka, and H. Abe (1977) Three-dimensional echocardiograms and two-dimensional echocardiographic images at desired planes by a computerized system. Ultrasound. Med. Biol.,
3:163-178.
McIntosh, J.R., K.L. McDonnald, M.K. Edwards, and B.M. Ross (1979)
Three-dimensional structure of the central mitotic spindle of diatome vulgare. J. Cell. Biol., 83:428-442.
Medina, F.J., M. Risueno, and S. Moreno de la Espina (1983) 3-D
reconstruction and morphometry of fibrillar centres in plant cells
in relation to nuclear activity. Biol. Cell., 48t31-38.
Medonca Dias, M.H., W.J. Mann, J. Chumas, M.L. Bernardo, and P.C.
Lauterbur (1982) Three-dimensional nuclear magnetic resonance
zeugmatographic imaging of surgical specimens. Bioscience Reports, 2t713-717.
Moens, P.B., and T. Moens (1981) Computer measurements and graphics of three-dimensional cellular ultrastructure. J. Ultrastruct. Res.,
78: 131-14 1.
Nierzwicki-Bauer,S.A., D.L. Balkwill, and S.E. Stevens (1983a) Threedimensional ultrastructure of a unicellular cyanobacterium. J. Cell
Biol., 97:713-722.
Nierzwicki-Bauer, S.A., D.L. Blakwill, and S.E. Stevens (198313)Use of
a computer-aided reconstruction system to examine the three-dimensional architecture of cyanobacteria. J. Ultrastruct. Res., 84:7382.
Nixon, J.V., S.I. Saffer, K. Lipscomb, and C.G. Blomqvist (1983) Threedimensional echoventriculography. Am. Heart J., 106t435-443.
Overdijk, J., H.B.M. Uylings, K. Kuypers, and A.W. Kamstra (1978)
An economical, semi-automatic system for measuring cellular tree
structures in three dimensions, with special emphasis on Golgiimpregnated neurons. J. Microsc., 114t271-284.
Paldino, A.M. (1979) A novel version of the computer microscope for
the quantitative analysis of biological structures: Applications to
neuronal morphology. Comp. Biomed. Res., 12:413-431.
Palmer, R., A. Yen, I. Kuo, D. Feinberg, P. Wiedenbeck, V. PerezMendez, C.G. Skioldebrand, and E. Carlsson (1982) Computer
graphics display of cardiac CT scans. Cardiovasc. Intervent. Radiol.
5:97-104.
Perkins, W.J., A.N. Barrett, J. Green, and D. Reynolds (1979) A system
for the three-dimensional construction, manipulation and display
of microbiological models. J. Biomed. Eng., 1:22-32.
Perkins, W.J., and R.J. Green (1982) Three-dimensional reconstruction
of biological sections. J. Biomed. Eng., 4:37-43.
Phong, B.T. (1975) Illumination for computer generated pictures. Commun. ACM., 18:311-317.
Prothero, J., and J. Prothero (1982) Three-dimensional reconstruction
from serial sections I. A portable microcomputer-based software
package in Fortran. Comp. Biomed. Res., 15:598-604.
Prothero, J.S., M. Riggins, A. Lindsay, R. Harris, and J.W. Prothero
(1985) Three-dimensional reconstruction from serial sections 111.
Autoscan, a software package in FORTRAN for semiautomated
photomicrography. Comp. Biomed. Res., 18:132-136.
Prothero, J., A. Tamarin, and R. Pickering (1974) Morphometrics of
living specimens. A methodology for the quantitative three-dimensional study of growing embryos. J. Microsc., 101:31-58.
Pykett, I.L., F.S. Buonanno, T.J. Brady, and J.P. Kistler (1983) True
three-dimensional nuclear magnetic resonance neuro-imaging in
ischemic stroke: Correlation of NMR, X-ray CT and pathology.
Stroke, 14t173-177.
Radermacher, M., and J. Frank (1984) Representation of three-dimensionally reconstructed objects in electron microscopy by surfaces of
equal density. J. Microsc., 136t77-85.
Rakic, P., L.J. Stensas, E.P. Sayre, and R.L. Sidman (1974) Computeraided three-dimensional electron microscopic montages of foetal
monkey brain. Nature, 250:31-34.
469
Reumann, K., G. Giebel, and K. Mildenstein (1985) Manufacturing
models of biomedical objects via CAD/CAM and GKS. Comput.
Graphics Forum, 4:375-382.
Ritman, E.L., J.H. Kinsey, R.A. Robb, B.K. Gilbert, L.D. Harris, and
E.H. Wood (1980) Three-dimensional imaging of heart, lungs and
circulation. Science, 210:273-280.
Ritman, E., R.A. Robb, S.A. Johnson, B.K. Gilbert, J.F. Greenleaf, R.E.
Sturm, and E.H. Wood (1978)Quantitative imaging of the structure
and function of the heart, lungs and circulation. MAY0 Clin. Proc.,
53t3-11.
Robb, R.A. (1982) Computer-based three-dimensional medical imaging.
J. Med. Syst., 6t535-537.
Robb, R.A., J.F. Greenleaf, E.L. Ritman, S.A. Johnson, J.D. Sjostrand,
G.T. Herman, and E.H. Wood (1974) Three-dimensional visualization of the intact thorax and contents: A technique for cross-sectional reconstruction from multiplanar X-ray views. Comp. Biomed.
Res., 7:395-419.
Robb, R.A., A.H. Lent, B.K. Gilbert, and A. Chu (1980) The Dynamic
Spatial Reconstructor: A computed tomography system for highspeed simultaneous scanning of multiple cross sections of the heart.
J. Med. Syst., 4:253-288.
Sadjani, F.A., and E.L. Hall (1980) Three-dimensional moment invariants. IEEE. PAMI., 2127-136.
Schierenberg, E., C. Carlson, and W. Sidio (1984) Cellular development
of a nematode: 3-D computer reconstruction of living embryos.
Rouxs Arch. Dev. Biol., 194%-68.
Serra, J. (1982) Image analysis and mathematical morphology. Academic Press, London.
Shannon, C.E. (19491, Proc. IRE., 37:lO-21.
Shantz, M.J., and G.D. McCann (1978) Computational morphology:
Three-dimensional computer graphics for electron microscopy.
IEEE. Trans. Biomed. Eng., 25:99-103.
Sinak, L.J., E.A. Hoffman, R.S. Schwartz, H.C. Smith, D.R. Holmes,
A.A. Bove, R.A. Robb, L.D. Harris, and E.L. Ritman (1985) Threedimensional cardiac anatomy and function in heart disease in
adults: Initial results with the Dynamic Spatial Reconstructor.
Mayo Clin. Proc., 6Ot383-392.
Sobel, I., C. Levinthal, and E.R. Macagno (1980) Special techniques for
the automatic computer reconstruction of neuronal structures.
Annu. Rev. Biophys. Bioeng., 9:347-362.
Stein, A., S. Juliano, P. Karp, and P. Hand (1984) Computer-assisted
radiographic images of the cerebral cortex. J. Neurosci. Methods,
IOt189-198.
Stevens, J.K., T.L. Davis, N. Friedman, and P. Sterling (1980) A systematic approach to reconstructing microcircuitry by electron microscopy of serial sections. Brain Res. Rev., 2t265-293.
Street, C.H., and R.R. Mize (1983) A simple microcomputer-based threedimensional serial section reconstruction system (MICROS). J.
Neurosci. Methods, 7:359-375.
Sullivan, P.G. (1983) Volume changes in the growing canine mandible
measured by intra-vital fluorescent dye three-dimensional reconstruction. Arch. Oral Biol., 28t482-489.
Sundsten, J.W., and J.W. Prothero (1983) Three-dimensional reconstruction from serial sections: 11. A microcomputer-based facility
for rapid data collection. Anat. Rec., 207r665-671.
Sutherland, I.E., R.F. Sproul, and R.A. Schumacker (1974) A characterization of ten hidden-surface algorithms. ACM Comp. Surv. 5.1-56.
Thompson, R.P., Y.M.M. Wong, and T.P. Fitzharris (1983) A computer
graphic study of cardiac truncal septation. Anat. Rec., 206t207214.
Totty, W.G., and M.W. Vannier (1984) Complex musculoskeletal anatomy: Analysis using three dimensional surface reconstruction. Radiology, 150:173-177.
Udupa, J.K. (1981) Determination of 3-D shape parameters from
boundary information. Comput. Graph. Image Proces., 17t52-59.
Udupa, J.K. (1982) Interactive segmentation and boundary surface
formation for 3-D digital images. Comput. Graph. Image Proces.,
18t213-236.
Udupa, J.K. (1983) Display of 3D information in discrete 3D scenes
produced by computerized tomography. Proc. IEEE, 71:420-431.
Van der Voort, H.T.M., G.J. Brakenhoff, J.A.C. Valkenburg, and N.
Nanninga (1985) Design and use of a computer-controlled confocal
microscope for biological applications. Scanning, 7:66-78.
Van Heel, M. (1983) Stereographic representation of three-dimensional
density distributions. Ultramicroscopy, 11:307-314.
Vannier, M.W., G.C. Conroy, J.L. Marsh, and R.H. Knapp (1985) Threedimensional cranial surface reconstructions using high-resolution
computed tomography. Am. J. Phys. Anthropol., 67r299-311.
Vannier, M.W., M.H. Gado, and J.L. Marsh 11983a) Three-dimensional
470
D.P. HUIJSMANS, W.H. LAMERS, J.A. LOS, AND J. STRACKEE
three dimensions for perspective reconstruction of brain ultrastrucdisplay of intracranial soft-tissue structures. AJNR 4520-521.
ture. IEEE Trans. Biomed. Eng., 20:288-291.
Vannier, M.W., J.L. Marsh, M.H. Gado, W.G. Totty, L.A. Gilula, and
R.G. Evens (1983b) Klinische Anwendungen der dreidimensiona- Wood, E.H. (1979) Noninvasive three-dimensional viewing of the motion and anatomical structure of the heart, lungs and circulation
len Overflachenrekonstruktion aus CT-Scans: Erfahrungen bei 250
system by high-speed computerized X-ray tomography. CRC Crit.
Untersuchungen. Electromedica, 51t122-131.
Rev. Biochem., 12161-186.
Vannier, M.W., J.L. Marsh, and J.O. Warren (1984) Three dimensional
CT reconstruction images for craniofacial surgical planning and Yasuda, T., J. Torikawa, S. Yokoi, and K. Katada (1984) A threedimensional display system of CT images for surgical planning.
evaluation. Radiology, 150:179-184.
IEEE ISMII Proc., pp. 322-327.
Veen, A., and L.D. Peachey (1977) TROTS A computer graphics system
for three-dimensional reconstruction from serial sections. Comput. Yelnik, J., G. Percheron, J. Perbos, and C. Francois (1981) A computeraided method for the quantitative analysis of dendritic arborizaGraph., 2r135-150.
tions reconstructed from serial sections. J. Neurosci. Methods,
Wann, D.F., T.A. Woolsey, M.L. Dierker, and W.M. Cowan (1973) An
4,347-364.
on-line digital-computer system for the semiautomatic analysis of
Golgi-impregnated neurons. IEEE Trans. Biomed. Eng., 4t233-247. Yokoi, S., J. Yorozu, S. Tsuruoka, and Y. Miyake (1983) A method for
three-dimensional display of CT image sequence. Medinfo 83. v.
Ware, R., and V. LoPresti (1975) Three-dimensional reconstruction
T.H. Bemmel, M.T. Ball, 0. Wigertz, eds. North-Holland, pp. 385from serial sections. Int. Rev. Cytol., 40:325-440.
388.
Willey, T.J., R.L. Schultz, and A.H. Gott (1973) Computer graphics in
Документ
Категория
Без категории
Просмотров
5
Размер файла
2 035 Кб
Теги
parallel, dimensions, morphometric, towards, generation, three, section, serial, package, aided, computerized, software, quantification, picture, review, computer, facilities, reconstruction
1/--страниц
Пожаловаться на содержимое документа