close

Вход

Забыли?

вход по аккаунту

?

Проект для научной выставки

код для вставкиСкачать
Presentation for WSC-8
The thermospectroscopic noninvasive glucometer
Sergey S. Krivenko
Anatolii A. Pulavskyi.
RPC “BioPromin”,Kharkov,Ukraine
Scope
The present document describes the detailed mapping a non-invasive easy-to-use blood glucose
meter. Diabetes mellitus is a chronic disease which is caused by deficit of insulin producing
by the pancreas or ineffective assimilation of the insulin by a body. There were 171 million
diabetics in the world as for 2000. 366 million diabetics were forecasted for 2030 [World
Health Organization].
2
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Abstract
The control of blood glucose concentration is a first-priority task, aimed at prevention of
complication, related to the consequences of diabetes mellitus. Existing methods of
controlling blood glucose concentration are invasive, i.e. requiring blood sampling (usually,
fingertip capillary blood). These methods have a number of restrictions, and main of them
is pain during carrying-out of an analysis. There is a non-invasive biochemical blood
analyzer AMP in medical practice, which uses the method of Professor Malykhin. It is
characterized by reliability, efficiency and acceptable accuracy. But this analizer uses five
spectral sensors and requires trained medical personnel to properly use. In addition, it is not
fully using the spectral information obtained from sensors.
3
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Objective of the work
The objective of our work is to create non-invasive home-used blood glucose meter that does
not require medical staff for its operation and need only two sensors. In addition, the
method uses the spectral information coming from sensors entirety. glucose meters.
4
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Agenda
• Methods for monitoring of blood glucose concentration
• Biological environment model
• The thermospectroscopic noninvasive glucometer prototype
• Results
• Conclusions
5
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
The structure of primary medical
care (according to WHO)
6
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Primary patient’s measures and invasive
methods disadvantages
Primary measures, aimed at
diagnosis
 blood glucose concentration
control
 blood pressure control
 chiropody
7
Disadvantages of invasive
methods
 painful during blood sampling
 possibility of a patient’s
infection
 daily fingertip puncture creates
nuisances in everyday life
 an average diabetic makes less
than 2 tests per day instead of
recommended 4-7 tests
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Light and skin
Interaction between light and
skin
8
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Thermally stabilized spectra of left
armpit
𝑻𝒆𝒓𝒎𝒐𝒔𝒑𝒆𝒄𝒕𝒓𝒖𝒎 = 𝒇(𝑺𝒑𝒆𝒄𝒕𝒓𝒖𝒎; 𝑯𝒖𝒎𝒊𝒅𝒊𝒕𝒚; 𝑻𝒆𝒎𝒑𝒆𝒓𝒂𝒕𝒖𝒓𝒆; … 𝒐𝒕𝒉𝒆𝒓𝒔 𝒔𝒆𝒏𝒔𝒐𝒓𝒔)
9
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Multivariate modeling
10
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Stages of developing mathematical model of
glucose concentration analysis
Preparation phase





clinical data collection – patients’ blood sampling
testing of glucose concentration in blood samples
measurement of thermo-stabilized spectra in the corresponding points
splitting of spectrum combination into two parts – calibration set and validation set
(duplex, Kennard-Stone method, histogram method)
spectra preprocessing
(Multiplicative scatter correction /standard deviation,
differentiating, nonlinear filtering)
11
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Stages of developing mathematical model of
glucose concentration analysis
Calibration phase







choice of a calibration method (PLS,RBF-PLS or ANN)
testing of glucose concentration in blood samples
model development in increasing order of complexity/increase in the number of factors
prediction of concentrations using developed models for calibration data set and
comparison of results with reference data (clinical analysis)
identification of outliers and bad values, and their removal
model recalibration with removed outliers
evaluation of root mean square error of calibration (RMSEC) and cross-validation of
each model. Formation of model complexity dependency on mean square error to define
optimal number of model factors (components)
12
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Stages of developing mathematical model of
glucose concentration analysis
Validation phase


prediction of concentrations using developed models for validation data set and
comparison of results with reference data (clinical analysis)model recalibration with
removed outliers
evaluation of root mean square error of prediction (RMSEP) for each model. Formation
of model complexity dependency on mean square error to define optimal number of
model factors (components)
13
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Neural Network Architecture Based on MLP
14
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
ANN or RBF-PLS?
RBF-PLS (B. Walczak et al.) works better than
classical 3-layer perceptron in OUR case. Why?
We don’t know.
We're just users, which use chemometrics.
15
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Experiment terms







Total patients' number – 238
Patients' age – 16-77 years
Patients belonged to Arab, Chinese, Slavic and Finno-Ugric groups
Diabetes: I and II types
Glucose concentration: 2-29 mmol/l
Calibration group – 166 patients, validation group – 72 patients
Hematocrit was not measured!
16
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Patients’ age histograms
17
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Glucose concentrations’ histograms
18
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Some comparison of results
19
PLS
RBF-PLS
R2
0.4
0.9
RMSEP
3.78
1.34
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Experiment's result. Table with some values
(according to RBF-PLS)
Glucose concentration (clinic analysis), mmol/l
10
10
5,5
5,5
5,44
7,2
15,6
15
5,5
5,5
4,2
4,2
5,3
4,3
5,2
5,2
5
5
7,1
7,1
5,1
6,5
5,1
6,6
6,3
20
8,9
Glucose concentration (model analysis), mmol/l
9,130446
9,942483
6,325168
5,429411
5,424803
6,255659
14,33417
10,79327
6,084031
6,087009
4,249031
4,193115
5,502326
3,69455
8,479128
5,508318
5,458402
5,138747
7,333616
7,357808
5,154192
6,444498
6,184803
6,567375
6,024245
WSC-8 Krivenko S.S., Pulavskyi8,982031
A.A.
Difference between clinical and model analysis
0,869554
0,057517
0,825168
0,070589
0,015197
0,944341
1,265825
4,206729
0,584031
0,587009
0,049031
0,006885
0,202326
0,60545
3,279128
0,308318
0,458402
0,138747
0,233616
0,257808
0,054192
0,055502
1,084803
0,032625
0,275755
9/23/2014
0,082031
Experiment's result. Clarke error grid
RMSEP=1.34; R2=0.9
21
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Experiment's result. Parkes error
grids for type I and II diabetics
22
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Experiment's result. ISO 15197:2003 tolerance
23
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Embedded calculate
RBF-PLS
Glucose
Sensors data
T=62,4 second
24
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Conclusions
A method for noninvasive evaluation of the concentration of glucose in human blood was
proposed. The advantages of this method is using of two sensors only, possibility of
frequent analysis, simplicity of operation by untrained user, which makes it suitable for selfdiagnosis. One of the best properties is the space robustness - the device is relatively
insensitive to the location of measurement points, that compensates the missing data by
timed analysis. Clinical analysis of the results showed their correctness – 100% of results lie
within tolerable limits of the generally accepted clinical error grids, 87.5% of recommended
95% comply with ISO 15197:2003.
25
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Acknowledgements
I want to thank
1) Oxana Rodionova – she helps me for 5 years. I’ve started
my chemometrics way with she. She is a beautiful woman.
2) Kim Esbensen – his book is my “bible” on chemometrics
3) Beata Walczak – RBF-PLS is cool approach. I like it.
4) WSC-8 – I feel a free breath with it
Thank you!
But…Could you write your question because of my English
пЃЉ
26
WSC-8 Krivenko S.S., Pulavskyi A.A.
9/23/2014
Документ
Категория
Презентации
Просмотров
12
Размер файла
1 705 Кб
Теги
1/--страниц
Пожаловаться на содержимое документа