The aim is to compute the growth rate of stem cells by using segmentation, feature extraction and pattern recognition which are the fundamental methods of digital image processing. DRLSE algorithm is applied for segmenting images. The DRLSE algorithm is an amalgamation of Canny Edge Detector algorithm and DRLSE method, which uses the four well potential function. Features are extracted from segmented images using GLCM method and finally Support Vector Machine (SVM) is used for pattern recognition and classification of stem cells.
COMPUTING THE GROWTH RATE OF STEM CELLS USING DIGITAL IMAGE PROCESSING
1. COMPUTING THE GROWTH RATE OF STEM
CELLS USING DIGITAL IMAGE PROCESSING
Batch Members:
M.Pratyusha
C.N.Priyadarshini
R.Sushmidha
Under the guidance of
Mrs.B.Pradeepa,M.E
Senior Assistant Professor
3. ABSTRACT
The influence and impact of digital image processing on modern society, science,
technology and art are tremendous and incredible. A stem cell considered to be
the undifferentiated biological unit cells is a kind of cell that can duplicate itself
over and over, providing new cells that can turn into cells with a specific
purpose. The stem cell analysis and research through the digital image
processing contributes its strong support to the doctors in the field of stem cell
research to yield the results in computerized technology rather than clinical
method. This kind of research through computerized method reduces the effects
on human body during transplantation and tends to be more time economical.
This project aims at computing the growth rate of stem cells by using
segmentation, feature extraction and pattern recognition which are the
fundamental methods of digital image processing. DRLSE algorithm is applied
for segmenting images. The DRLSE algorithm is an amalgamation of Canny
Edge Detector algorithm and DRLSE method, which uses the four well potential
function. Features are extracted from segmented images using GLCM method
and finally Support Vector Machine (SVM) is used for pattern recognition and
classification of stem cells.
4. INTRODUCTION
ο STEM CELL - DEFINITION
ο SOURCES OF STEM CELLS
ο POSSIBLE USES OF STEM CELLS
ο FLOW CHART
5. STEM CELL- DEFINITION
ο Undifferentiated cells that can differentiate into specialized
cells and divide to produce more stem cells
ο Through cell division-form more stem cells
οEg: Bone marrow
ο Become specialized cells under critical condition
οEg: Pancreas & Heart
6. SOURCES OF STEM CELL
ο Umbilical cord blood
ο Bone marrow
ο Peripheral blood
ο Human embryos
7. ADVANTAGES OF STEM CELL
ο Replaceable tissues/organs
ο Repair of defective cell types
ο Delivery of genetic therapies
ο Delivery chemotherapeutic agents
8. FLOW CHART
INPUT IMAGE
(IMAGES FROM TIME LAPSE
VIDEO)
SEGMENTATION
(CANNY & 4 WELL DRLSE)
FEATURE EXTRACTION
(GRAY LEVEL CO
OCCURRENCE MATRIX)
PATTERN RECOGNITION
(SUPPORT VECTOR MACHINE)
9. CANNY EDGE DETECTOR
ο Developed by JOHN F.CANNY in 1986
ο Called optimal edge detector β satisfies three criterion:
οLow Error Rate
οLocalized edges
οSingle edge response
10. CANNY-ALGORITHM
ο Apply Gaussian filter to smooth the image in order to remove
the noise
ο Find the intensity gradients of the image
ο Apply non-maximum suppression to get rid of spurious
response to edge detection
ο Apply double threshold to determine potential edges
ο Track edge by hysterises
ο In matlab canny is applied by using the keyword
e = edge (I,βcannyβ);
11. DRLSE METHOD
ο Developed by C. Li, C. Xu, C. Gui, and M. D. Fox
ο It uses edge-based active contour method to drive level set
function in the desired
12. FOUR WELL POTENTIAL FUNCTION
ο The four well potential function is aimed to maintain the
signed distance property.
ο The four well potential is used to increase the quality of
segmented image and get better accuracy.
13. EVALUATION OF SEGMENTATION
The evaluation of segmentation is carried out using the following
parameters:
ο PSNR
ο BDE
ο PRECISION
ο RECALL
14. PSNR
PSNR = 10.πππ10
πβ2552
π π πΈ ππβπ ππ
2
Where N=SIZE OF IMAGE
O= ORIGINAL IMAGE
E= SEGMENTED IMAGE
MSE =
1
π π π πΈππ β πππ
2
where
N = SIZE OF IMAGE
O= ORIGINAL IMAGE
E= SEGMENTED IMAGE
15. PRECISION & RECALL
ο PRECISION:
=
π‘π
π‘π+ππ
ο RECALL:
=
π‘π
π‘π+ππ
Where
tp = intersection of segmented parts and ground truth
fp= segmented parts not overlapping ground truth
fn = missed parts of the ground truth
16. BOUNDARY DISPLACEMENT ERROR
The Boundary Displacement Error (BDE) measures the
average displacement error of one boundary pixels and the
closest boundary pixels in the other segmentation.
π πΏπ΄( π’, π£)= u-v/L-1 for 0<u-v<L
0 for u-v<0
16
17. COMPARISON OF CASE 1 AND CASE 2
13.6
13.8
14
14.2
14.4
14.6
14.8
15
0 hours 8 hours 16 hours 24 hours 32 hours 40 hours 48 hours 56 hours
BDE
WITH CANNY
WITHOUT CANNY
22. FEATURE EXTRACTION
ο Feature extraction is a special form of dimensionality
reduction
ο GLCM is a well-established statistical method for feature
extraction
οComputing the co-occurrence matrix
οCalculating feature based on the co-occurrence matrix
23. GLCM
ο In 1973, Haralick introduced the co-occurrence matrix and
texture features
ο Matrix is square with dimension Ng, where Ng is the number
of gray levels in the image
ο Features used are Auto correlation, contrast, dissimilarity,
homogeneity, energy and entropy
26. FEATURE EXTRACTION OUTPUT
FEATURES IMAG
E 1
IMAG
E 2
IMAG
E 3
IMAG
E 4
IMAG
E 5
IMAG
E 6
IMAG
E 7
IMAG
E 8
AUTO
CORRELATI
ON
1.1833 1.2014 1.218
1
1.2705 1.3191 1.3701 1.3668 1.3880
CONTRAST 2.4170 2.8029 2.638
1
2.9597 2.6180 4.0452 49899 5.5527
ENTROPY 3.3671 3.4131 3.760
6
4.3084 4.5614 5.3890 56441 5.9454
ENERGY 8.5460 8.4835 8.319
8
7.9916 7.7754 7.3479 7.2549 7.8074
DISSIMLARI
TY
2.4170 2.8029 2.638
1
2.9597 2.6180 4.0452 4.9899 5.5527
HOMOGENI
TY
9.8791 9.8958 9.868
0
9.8520 9.8690 9.7977 9.7505 9.7223
27. PATTERN RECOGNITION
ο Pattern recognition is the process of classifying input data into
objects or classes based on key features
ο SVMs introduced in COLT-92 by Boser, Guyon & Vapnik
31. REFERENCES
[1] Nuseiba M.Altarawneh,Suhuai Luo,Brian Regan and
Changming Sun, βA Modified Distance Regularized Level Set
Model for Liver Segmentation from CT Images,β Signal & Image
Processing : An International Journal(SIPIJ), vol.6,No.1,February
2015.
[2] Arathi J.Vyavahare, βCanny based DRLSE Algorithm for
Segmentation,β International Journal of Computer Applications,
vol.102-No.7,September 2014.
[3] C.Xu,d.l.pham,and j.1.prince,βMedical Image Segmentation
Using Deformable Models,β in SPIE Handbook on Medical
Imaging vol.3,J.M.Fitzpatrick and M.Sonka,Eds.,ed,2000,pp.129-
174.
32. CONTDβ¦..
[4] Punam Thakare,βA Study of Image Segmentation and Edge
Detection Techniques,β International Journal on Computer
Science and Engineering,vol 3, no.2 89-904,2011.
[5] Chuming Li,Chenyang Xu,βDistance Regularized Level Set
Evolution and its Application to Image Segmentationβ, IEEE
Trans on Image Processing,vol 19,no 12,December 2012.
[6] Jiafu Jiang , He Wei , Qi Qi ,β Medical Image Segmentation
Based on Biomimetic Pattern Recognitionβ, World Congress on
Software Engineering,IEEE Computer Society. Vol:2,Pp.375-
379,2009.
From this we can conclude tht the 2 striking features of stem cells is : self renewal & potency. Based on potency stem cell divided into typesβ¦given in slide 4
Bone marrow-tissue making blood cells,used whn umbilical sc not present.umbilical-taken from cord after baby born. Peripheral blood-done for BMT,sc separated from blood (donor) & blood returned to patients & sc injected. From aborted tissues sc taken-ethical issues.
Edge detection refers to process of identifying & locating sharp discontinuities=abrupt changes in pixel intensity.1.accurate detection of edges(as many as possible),2.dist.b/w edges found by detector & actual edge min,3.no false edge detection
1.Gaussian filter uses simple mask,performed using simple convolution method,larger the width lower is the sensitivity to noise,2.uses 4 dect to compute edges(hori,vertical=sobel; diag=roberts),sobel uses 3*3 mask,mag or edge strength & direction is calculated; 3.edge thinning,in the name itself,comparison of pixel with 1 abve & below or 90,180,135,0; 4.still some noise present,T1 & T2; 5.weak edges=either by noise or from true edge,blob analysis (N8)
Advantage:guide the direction of evolving contour
Mean square error (MSE) indicates the average difference of the pixels throughout the image. A higher MSE indicates a greater difference between the original and processed image. The PSNR computes the peak signal-to-noise ratio between two images, in decibels. This ratio is often used as a quality measurement between the original and a resultant image. The higher PSNR, the better the quality of the output image. To compute the PSNR , mean-squared error calculated is used
precision is the proportion of boundary pixels in the automatic segmentation that correspond to boundary pixels in the ground truth. Precision and recall are attractive as measures of segmentation quality because they are sensitive to over and under-segmentation, over-segmentation leads to low precision scores, while under-segmentation leads to low recall scores.
Recall is defined as the proportion of boundary pixels in the ground truth that were successfully detected by the automatic segmentation
The main goal ofΒ feature extractionΒ is to obtain the most relevant information from the original data and represent that information in a lower dimensionality space. When the input data to anΒ algorithmΒ is too large to be processed and it is suspected to be redundant (e.g. the same measurement in both feet and meters, or the repetitiveness of images presented asΒ pixels), then it can be transformed into a reduced set ofΒ featuresΒ (also named aΒ feature vector).
In 1973, Haralick introduced the co-occurrence matrix and texture features which are the most popular second order statistical features today.
From the wordβ¦it can be said tht it is a matrix.
Each element (i, j) in GLCM specifies the number of times that the pixel with value i occurred horizontally adjacent to a pixel with value j .
It works really wellΒ withΒ clear margin of separation
It is effective in high dimensional spaces.
It is effective in cases where number of dimensions is greater than the number of samples.
It uses a subset of training points in the decision function (called support vectors), so it is also memory efficient.
From the wordβ¦it can be said tht it is a matrix.
Each element (i, j) in GLCM specifies the number of times that the pixel with value i occurred horizontally adjacent to a pixel with value j .
Cons:
ItΒ doesnβt perform well, when we have large data set because the required training time is higher
It also doesnβt perform very well, when the data set has more noise i.e. target classes are overlapping
SVM doesnβt directly provide probability estimates, these are calculated using an expensive five-fold cross-validation. It is related SVC method of Python scikit-learn library.