Thesis presentation

72

description

 

Transcript of Thesis presentation

Page 1: Thesis presentation
Page 2: Thesis presentation

CLASSIFICATION OF DIABETIC MACULAR EDEMA USING COLORED

FUNDUS IMAGE

PRESENTED BY:

MUHAMMAD ZUBAIR

SUPERVISED BY:

Dr. SHOAB AHMED KHAN

Page 3: Thesis presentation

GUIDANCE AND EVALUATION COMMITTEE

GEC Members

Dr. Ubaid Ullah Yasin (Co-Supervisor)(Asst. Prof. AFPGMI, Eye specialist

AFIO)

Dr. Aasia Khanum

Dr. Ali Hassan

Page 4: Thesis presentation

AGENDA Motivation and problem statement

Introduction

Literature review

Proposed System

Evaluation and Results

Conclusion and Future work

References

Page 5: Thesis presentation

MOTIVATION AND PROBLEM STATEMENT

Page 6: Thesis presentation
Page 7: Thesis presentation

MOTIVATION

Diabetes Mellitus is a fast growing global disease

In 2007 survey by International Diabetic Federation (IDF) Pakistan was at 7th in top 10 countries having diabetic population [1]

This number may rise from 6.9 million in 2007 to 11.5 million in 2025 by IDF [1], almost double

Page 8: Thesis presentation

STATISTICAL CHART BY IDF2007 2025

Country Persons (million

s)

Country Persons (millions

)

1 India 40.9 1 India 69.9

2 China 39.8 2 China 59.3

3 USA 19.2 3 USA 25.4

4 Russia 9.6 4 Brazil 17.6

5 Germany 7.4 5 Pakistan 11.5

6 Japan 7.0 6 Mexico 10.8

7 Pakistan 6.9 7 Russia 10.3

8 Brazil 6.9 8 Germany 8.1

9 Mexico 6.1 9 Egypt 7.6

10

Egypt 4.4 10

Bangladesh 7.4

Page 9: Thesis presentation

MOTIVATION … DME can lead to complete irreversible

blindness

Early stage detection is rare

Doctor to patient ratio is very low in Pakistan

Screening is hectic in populated areas

Sparse resources in rural areas

Page 10: Thesis presentation

PROBLEM STATEMENT To develop an automated CAD system to help

ophthalmologists in mass screening by

Identification of abnormalities (exudates),

Finding out their exact location and area within the macular region,

Stage classification of the disease

Page 11: Thesis presentation

PUBLICATIONS CONFERENCES

“Classification of Diabetic Macular Edema and Its Stages Using Color Fundus Image” (Registered and to be presented in 2013 3rd International conference on Signal, Image Processing and Applications (ICSIA))

“Automated Detection of Optic Disc for the Analysis of Retina Using Color Fundus Image” (Accepted in 2013 IEEE International conference on Imaging Systems And Techniques (IST))

“Automated Segmentation of Exudates Using Dynamic Thresholding in Retinal Photographs” (Accepted in 16th IEEE International Multi Topic Conference (INMIC) 2013)

JOURNALS “Automated Grading of Diabetic Macular Edema Using Colored

Fundus Photographs” (Submitted in Journal of Digital Imaging)

Page 12: Thesis presentation

INTRODUCTION

Page 13: Thesis presentation

HUMAN EYE

Retinal layer of human eye comprises:

o Optic Disc (Head of optic nerve)

o Macula (Central portion of retina)

o Fovea (Center of macula)

o Network of Blood vessels

Page 14: Thesis presentation

Blood Vessels

Macula

Fovea

Optic Disc

COLORED FUNDUS IMAGE

Page 15: Thesis presentation

DIABETIC MACULAR EDEMA

DME is the swelling of the macula

Leaked fluid accumulates on the retina

Leakage within the macula causes swelling

Page 16: Thesis presentation

SCREENING TESTS FOR DME

Fundus Fluorescein Angiography (FFA)

Retinal image acquisition process

2D colored and red free images

Optical Coherence Tomography (OCT)

3D high resolution image acquisition process

Provides a cross-sectional image

Used as optical biopsy by the ophthalmologists

Page 17: Thesis presentation

FFA IMAGE ACQUISITION PROCESS

Page 18: Thesis presentation

FFA COLORED IMAGE

Page 19: Thesis presentation

OCT IMAGE ACQUISITION PROCESS

Page 20: Thesis presentation

OCT IMAGE

Page 21: Thesis presentation

LITERATURE REVIEW

Page 22: Thesis presentation

Aquino et al.*

Detect OD using edge detection

Circular Hough transform

Feature extraction

Morphological operations

Accuracy achieved was 86%

A. Aquino, M. E. G. Arias and D. Marin, “Detecting the Optic Disc Boundary in Digital Fundus Images Using Morphological, Edge Detection and Feature Extraction Techniques,” IEEE TRANSACTIONS ON MEDICAL IMAGING, Vol. 29, No. 11, pp.1860-1869, 2010

Page 23: Thesis presentation

Siddalingaswamy et al.*

Exudates detection using clustering

Morphological techniques

Exudates location based severity level

Sensitivity achieved 95% and specificity 96%

No publically available database used

Use only 148 local fundus images

P.C. Siddaligaswamy, K. G. Prabhu, “Automatic Grading of Diabetic Maculopathy Severity Levels”, Proceedings of 2010 International Conference on Systems in Medicine and Biology, pp. 331-334, 2010

Page 24: Thesis presentation

Lim et al.*

Segmentation of exudates using watershed transform

External and internal markers used

Classification done as normal, stage1 and stage2

Use only 88 images of MESSIDOR

Sensitivity and Specificity achieved 80.9%, 90.2% respectively

Accuracy achieved was 85.2 percent

S.T. Lim, W.M.D.W. Zaki, A. Hussain, S.L. Lim, S. Kusalavan, “Automatic Classification of Diabetic Macular Edema in Digital Fundus Images", 2011 IEEE Colloquium on Humanities, Science and Engineering (CHUSER), pp. 265-269, 2011.

Page 25: Thesis presentation

Giancardo et al.*

Unsupervised technique for exudates segmentation

Background estimation used

Image normalization used

8 neighbor connectivity used

Used local database images

No Stage classification was done

L. Giancardo, F. Meriaudeau, T.P. Karnowski, K. W. Tobin Jr, E. Chaum, MD, “AUTOMATIC RETINA EXUDATES SEGMENTATIONWITHOUT A MANUALLY LABELLED TRAINING SET”, IEEE Transactions on Medical Imaging, Vol. 21, No. 5, pp. 1396-1400, 2011

Page 26: Thesis presentation

PROPOSED SYSTEM

Page 27: Thesis presentation

PROPOSED SYSTEM

Preprocessing of image

Optic disc elimination

Dynamic Thresholding

All possible exudates detection

Classification of the stage

Page 28: Thesis presentation

FLOW CHART OF PROPOSED TECHNIQUE

Input Colored Fundus Image

Resizing of Image

Green Component

Preprocessing

OD Removal

Fovea localization

Grayscale

Classifier

Page 29: Thesis presentation
Page 30: Thesis presentation

Input Colored Image

Green Channel of Image

Contrast Limited Adaptive Histogram Equalization

Contrast Stretching Transform

PREPROCESSING OF IMAGE

Page 31: Thesis presentation

PREPROCESSING OF IMAGECLAHE

Contrast Limited Adaptive Histogram Equalization (CLAHE)

To make the components visually distinct

Distinction of foreground objects from background

CLAHE used iteratively to get the desired result

Page 32: Thesis presentation

Colored image

1st CLAHE

2nd CLAHE

Green component

Page 33: Thesis presentation

PREPROCESSING OF IMAGECONTRAST STRETCHING TRANSFORM

Expands the range of intensity of the image

Range span is defined by the user

Improves the overall contrast of the image

Increases the contrast b/w dark and lights

Stretches the intensity domain histogram

Page 34: Thesis presentation

Colored image

1st CLAHE

2nd CLAHE

Green component

Contrast Stretching

Page 35: Thesis presentation

RESULTS OF PREPROCESSING

Green Component

1st CLAHE 2nd CLAHE

Contrast Stretching

Page 36: Thesis presentation

OPTIC DISC (OD) ELIMINATION OD becomes prominent after preprocessing

Detection on basis of highest intensity value pixels

Extended minima transform applied

Detect all candidate OD regions

Morphological operations to extract real OD

Page 37: Thesis presentation

OD EXTRACTION FLOW CHART

Page 38: Thesis presentation

OPTIC DISC (OD) ELIMINATIONEXTENDED MINIMA TRANSFORM

Computes the regional minima

Regional minima are connected components of pixels with constant intensity value

The external boundary pixels have high value

8-neighbor connectivity used

Selects all possible candidate regions for OD

Page 39: Thesis presentation

8 neighbor connectivity

(x-1 , y+1) (x , y+1) (x+1 , y+1)

(x-1 , y) (x , y) (x+1 , y)

(x-1 , y-1) (x , y-1) (x+1 , y-1)

Page 40: Thesis presentation

Contrast Stretching Transform

Extended minima Transform

EXTENDED MINIMA TRANSFORM RESULT

Page 41: Thesis presentation

MORPHOLOGICAL OPERATIONS Morphological Erotion

To remove all non OD candidate regions

Structuring element (SE) having size less than OD size

SE chosen is of the size one fourth of the radius of OD

Erotion also shrinks the actual OD region

Page 42: Thesis presentation

MORPHOLOGICAL OPERATIONS Morphological Dilation

To get the actual size of OD after erotion

SE is of the size approx. equal to the radius of OD

Dilation restores the size and shape of actual OD

False OD are no more in the image

Page 43: Thesis presentation

PREPROCESSING AND OD ELIMINATION

Page 44: Thesis presentation

RESULTS OF OD ELIMINATION

Page 45: Thesis presentation

DYNAMIC THRESHOLDING Fundus images are taken in different

illumination

All images have different intensity levels

Intensity based parameters chosen for dynamic thresholding

Mean and standard deviation of the image is used

Threshold value is set using these parameters

Page 46: Thesis presentation

CALCULATION OF THRESHOLD VALUE USING MEAN AND STANDARD DEVIATION

Mean Standard deviation

Threshold value

30.05 33.35 2.831.25 34.6 2.831.25 35.4 2.831.9 35.7 2.9

32.45 36.2 2.932.55 35.92 3.032.84 35.96 3.035.08 38.6 3.235.09 40.2 3.335.49 39.38 3.335.85 39.40 3.335.95 39.46 3.337.15 41.87 3.437.35 41.56 3.438.57 42.27 3.441.00 46.65 3.641.22 47.1 3.6541.53 47.06 3.6543.10 48.48 3.743.07 48.05 3.7543.27 47.71 3.7544.58 49.58 3.945.56 51.30 3.9548.20 53.50 4.349.4 54.51 4.4

Page 47: Thesis presentation

2.8

3.0

3.9

3.6

4.4

4.1

3.4

Page 48: Thesis presentation

DETECTION OF EXUDATES

Exudates are bright lesions

Bright yellowish spots in colored fundus image

OD free image is used for exudates detection

Segmentation is done for exudates detection

Page 49: Thesis presentation

RESULTS OF EXUDATES DETECTION

Page 50: Thesis presentation

ETDRS STAGING CRITERIA Early Treatment Diabetic Retinopathy Study

(ETDRS)

Classification of DME is based on the standard criteria set by ETDRS

Severity level depends upon size and location of abnormality

Reference point is the center of macula

Page 51: Thesis presentation

CIRCULAR AREA FOR STAGING

Page 52: Thesis presentation

RESULTS OF EXUDATES DETECTION

Page 53: Thesis presentation

CLASSIFICATION OF STAGES

Normal Condition No abnormality is found

Less Significant Stage Abnormality found outside the circular area of 1 Disc

Diameter 1DD radius but within 2DD from the fovea

Moderate Stage Abnormality found beyond 1/3DD (radius) but

within 1DD of circle

Severe Stage Abnormality found within 1/3DD circular area from

the center of fovea region

Page 54: Thesis presentation

STAGE CLASSIFICATION TABLE

Stage Description

Normal No abnormality is found

SevereAbnormality found within 1/3DD radius circular area from the center of fovea region

Moderate Abnormality found beyond 1/3DD radius but within 1DD radius of circle

Less significant

Abnormality found outside the circular area of 1DD radius but within 2DD radius from the fovea region

1Disc Diameter (DD) = 1500µm (1.5mm)

Page 55: Thesis presentation

CLASSIFICATION OF STAGES

Colored input image

Normal Case

Colored input image

Less Significant Stage

Page 56: Thesis presentation

Colored input image

Colored input image

Moderate Stage

Severe Stage

CLASSIFICATION OF STAGES…

Page 57: Thesis presentation

RESULTANT MESSAGES DISPLAYED AFTER THE CLASSIFICATION OF STAGES

If no exudates are found in the image or the abnormality is found outside the 2DD circle, it is declared as normal case. The message will be displayed “No exudates found: Normal eye”.

If the exudates are found within the outermost 2DD circle the stage is called insignificant. The message will be displayed “warning: exudates found within the outermost circle, insignificant stage”.

If the exudates are found in both outermost 2DD and the middle circle 1DD circle, it is declared as moderate stage. The message will be displayed “warning: exudates found within the outermost circle and middle circle, moderate stage”.

If the exudates are found in the middle circle only 1DD circular area it is known as moderate stage. The message will be displayed “warning: exudates found within the middle circle, moderate stage”.

If the abnormality (exudates) found within both the middle circle 1DD circular area and the innermost 1/3DD circle the stage is known as severe stage. The message will be displayed “warning: exudates found within the middle circle and the innermost circle, severe stage”.

If the exudates are found within the innermost 1/3DD circle the stage is called severe stage. The message displayed in this case “warning: exudates found in the innermost circle, severe stage”.

Page 58: Thesis presentation

EVALUATION AND RESULTS

Page 59: Thesis presentation

RESULTS Performance metrics

SensitivitySpecificityAccuracy

Sensitivity = TP / (TP + FN)

Specificity = TN / (TN + FP)

Accuracy = (TP+ TN) / (TP + FN + TN + FP)

TP for True Positive (Correctly Identified) TN for True Negative (Correctly Rejected) FP for False Positive (Incorrectly Identified) FN for False Negative (Incorrectly Rejected)

Page 60: Thesis presentation

CLASSIFICATION RESULTS

Total no. of images

Normal

Cases

Insignificant

Moderate Severe

1200 869 122 136 73

Proposed Method

Sensitivity (%)

Specificity (%)

Accuracy (%)

98.27 96.58 96.54

Page 61: Thesis presentation

EXUDATES DETECTION RESULT

Author Sens(%) Spec(%) Accu(%)

Deepak et al. 100 97 81

Giancardo et al. - - 89

Proposed technique

98.73 98.25 97.62

Page 62: Thesis presentation

GRAPHICAL RESULT FOR EXUDATES DETECTION

Deepak et al Giancardo et al. Proposed technique

0

20

40

60

80

100

120

SensitivitySpecificityAccuracy

Page 63: Thesis presentation

AuthorSensitivity

(%)Specificity

(%)Accuracy

(%)

Lim et al. 80.9 90.2 85.2

Deepak et al. 95 90 -

Proposed Method

98.27 96.58 96.54

CLASSIFICATION OF DME RESULT

Page 64: Thesis presentation

GRAPHICAL RESULT FOR CLASSIFICATION OF DME

Lim et al. Deepak et al. Proposed technique0

5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

SensitivitySpecificityAccuracy

Page 65: Thesis presentation

CONCLUSION AND FUTURE WORK

Page 66: Thesis presentation

CONCLUSION Improved performance of the system in terms of

True localization of optic disc and its removal Detection of exudates and determining their position Classifying stage of the disease

The System provide accurate results using less but effective features of the input image

Efficient system for identifying and classifying the DME

The system can be used practically in diagnostic environment with a sound reliability

Page 67: Thesis presentation

FUTURE WORK Classification of DME using OCT (3D) images

Fusion of FFA and OCT results for easy and better diagnosis

Detection of soft exudates/drusens in fundus images

Page 68: Thesis presentation

REFERENCES

Page 69: Thesis presentation

REFERENCES [1] International Diabetic Federation Atlas. 2006 showing prevalence of

diabetes in 2007 and future projection for 2025. Available from: http://www.idf.org/diabetesatlas. (Updated in 2012)

[2] A. Aquino, M. E. G. Arias and D. Marin, “Detecting the Optic Disc Boundary in Digital Fundus Images Using Morphological, Edge Detection and Feature Extraction Techniques,” IEEE TRANSACTIONS ON MEDICAL IMAGING, Vol. 29, No. 11, pp.1860-1869, 2010

[3] K. S. Deepak and J. Sivaswamy, “Automatic Assessment of Macular Edema From Color Retinal Images", IEEE Transactions on Medical Imaging, Vol. 31, No. 3, pp. 766-776, 2012

[4] P.C. Siddalingaswamy, K. G. Prabhu, “Automatic Grading of Diabetic Maculopathy Severity Levels”, Proceedings of 2010 International Conference on Systems in Medicine and Biology, pp. 331-334, 2010

[5] L. Giancardo, F. Meriaudeau, T.P. Karnowski, K. W. Tobin Jr, E. Chaum, MD, “AUTOMATIC RETINA EXUDATES SEGMENTATION WITHOUT A MANUALLY LABELLED TRAINING SET”, IEEE Transactions on Medical Imaging, Vol. 21, No. 5, pp. 1396-1400, 2011

[6] Klein R, Klein BE, Moss SE, Davis MD, DeMets DL. The Wisconsin epidemiologic study of diabetic retinopathy. IV. Diabetic macular edema Ophthalmology 1984; 91(12): 1464-74

Page 70: Thesis presentation

REFERENCES… [7] G.S. Annie Grace Vimala, S. Kaja Mohideen, “Automatic Detection of Optic Disc

and Exudate from Retinal Images Using Clustering Algorithm,” Intelligent Systems and Control (ISCO), pp.280-284, Jan. 2013

[8] Shijian Lu, “Automatic Optic Disc Detection using Retinal Background and Retinal Blood Vessels,” Biomedical Engineering and Informatics (BMEI), vol.1, pp.141-145, Oct. 2010

[9] T. McInerney and D. Terzopoulos, T-snakes: Topology adaptive snakes", Med Image Anal., Vol. 4, pp. 73-91, 2000

[10] Hoover, V. Kouznetsova and M. Goldbaum, “Locating blood vessels in retinal images by piecewise threshold probing of a matched filter response”, IEEE Trans Med Imag., Vol. 19, pp. 203-211, 2000

[11] Sumathy.B, and Dr Poornachandra S, “Retinal blood vessel segmentation using morphological structuring element and entropy thresholding”, Conf. on Computing Communication & Networking Technologies (ICCCNT), pp.1-5, July 2012

[12] S. Sekhar, W.A. Nuaimy, A.K. Nandi, “Automatic Localisation of optic disc and fovea in retinal images", 16th European Signal Processing Conference (EUSIPCO), 2008

[13] Asim, Khawaja Muhammad, A. Basit, and Abdul Jalil. "Detection and localization of fovea in human retinal fundus images" In Emerging Technologies (ICET), 2012 International Conference on, pp. 1-5. IEEE, 2012

[14] U. R. Acharya, C. K. Chua, E. Y. K. Ng, W. Yu, C. Chee, “Application of Higher Order Spectra for the Identification of Diabetes Retinopathy Stages", Journal of Med Systems Vol. 32, pp. 481-488, 2008

Page 71: Thesis presentation

REFERENCES… [15] K. S. Deepak and J. Sivaswamy, “Automatic Assessment of Macular

Edema From Color Retinal Images", IEEE Transactions on Medical Imaging, Vol. 31, No. 3, pp. 766-776, 2012

[16] Jack J Kanski and Brad Bowling, “Clinical Ophthalmology A Systematic Approach,” seventh edition, May 2011

[17] MESSIDOR: http://messidor.crihan.fr/index-en.php (visited: 19 Feb 2013) [18] Xiaolu Zhu and Rangaraj M. Rangayyan, “Detection of the Optic Disc in

Images of the Retina Using the Hough Transform,” IEEE Conf. on Engineering in Medicine and Biology Society, pp.3546-3549, Aug. 2008

Page 72: Thesis presentation