Image Segmentation Project

1. FINAL PROJECT REPORT
Automatic Classification of Different Geometries of
Drusen in patients with Age related Macular
Degeneration
-Darshan Shah

2. PROJECT OVERVIEW
 What is the project about?

3. PROJECT OVERVIEW
 GOAL
 The goal of this project is to create a GUI to
automatically segment Drusen and categorize the
Drusen into its various types
Image courtesy harbour.com

4. PROJECT OVERVIEW
 SCOPE
 This project will help the eye examiners to find Drusen in the
image and help save time in categorizing Drusen thereby
speedingHomogeneity (Comparing area between
up the examination of the retina.
RPE and Choroid) Homogeneity
(Comparing area between RPE and
Choroid)
Image courtesy ophmanagement.com

5. LITERATURE REVIEW
KEY PAPER
Drusen Ultrastructure Imaging with Spectral Domain Optical
Coherence Tomography in Age-related Macular Degeneration -
Khanifar Aziz et al 2008
Quantitative Comparison of Drusen Segmented on SD-OCT versus
Drusen delineated on Color Fundus Photographs- Nieraj Jain et al
2010
Automatic segmentation of seven retinal layers in SDOCT images
congruent with expert manual segmentation - Stephanie J. Chiu
2010

7. HOMOGENEITY (COMPARING AREA
BETWEEN RPE AND CHOROID)
 Homogeneous – compare the entire layer for uniformity
 Non-homogeneous with a central core – varying
internal reflectivity with a distinct single focus of hyper-
reflectivity
 Non-homogeneous without a central core– varying
internal reflectivity

8. ORIGINAL LAYERED IMAGE

9. REGION OF INTEREST

10. SEGMENTED REGION

11. ATTEMPTED METHOD 1
 Taking max values along the vertical gradients

12. ATTEMPTED METHOD 2
 Convolving with a binary mask depending on
weights

13. METHOD 3
 GRAPH THEORY
 Implementation of Stephanie Chiu’s algorithm
 Using Dijkstra Algorithm to implement
lowest weighted path

14. SHIFTED AND SEGMENTED

15. AUTOMATIC INITIALIZATION

16. SO GETTING BACK TO SEGMENTED

17. DATA SET OF UNSEGMENTED IMAGES
mean before Std before
Image no Separate layer exists? segmentaion segmentation
51 yes 162.0504 35.2787
52 yes 156.1071 39.2795
53 yes 164.7804 38.8632
54 yes 162.8016 37.2076
55 yes 163.5681 35.8284
56 yes 172.4292 39.5759
57 yes but hardly 182.4318 37.7275
58 yes but hardly 185.2433 31.3095
59 yes but hardly 178.0332 33.8844
60 yes but hardly 167.4353 34.3635
61 not really 179.9425 35.9614
62 really bright 177.7636 34.5737
63 not really 181.2146 36.3007
64 separable diff with RPE 172.3632 31.1072
65 not really 179.5103 31.5538
66 no 183.3249 32.8943
67 not really 179.101 34.1617
68 no 184.9984 31.9882
69 no 184.9699 34.9764
70 no but some pecularity 183.9566 34.4765

18. BASED ON THIS ANALYSIS
A cut-off value for applying the
segmentation was decided
If the values was above the cut off it was
not segmented and the software proceeds
with classification directly
RUN CODE

19. RESOURCES
Literature
Paper of
Khanifar
Reference
Matlab
GUI code
Project
GUI Web
handbook resource

20. RFERENCES
 Toth CA, Farsiu S, Khanifar AA, Chong G, Application of Spectral Domain OCT in AMD,
in Optical Coherence Tomography in Age-Related Macular Degeneration, Gabriel
Coscas (Editor), (Springer Medizin Verlag Heidelberg, 2009), pp. 15-34:
 S.J. Chiu, X.T. Li, P. Nicholas, C.A. Toth, J.A. Izatt, S. Farsiu. “Automatic Segmentation
of Seven Retinal Layers in SDOCT Images Congruent with Expert Manual
Segmentation.” Optics Express. Vol. 18, No. 18, pp. 19413-19428, August 2010.
 N. Jain, A. A. Khanifar, S. Farsiu, S. Bearelly, S. J. Chiu, K. P. Winter, J. A. Izatt, and C.
A. Toth, Qualitative Comparison of Segmented Drusen on SD OCT versus Drusen
Delineated on Color Fundus Photograph Presented at the Association for Research in
Vision and Ophthalmology (ARVO) Annual Meeting, Fort Lauderdale, FL. May 2009.
 Dijktra code by Joseph Kirk-matlab mathworks
 Adjancey matrix code by Kevin Murphy - matlab mathworks

21. QUESTIONS?

22. APPENDIX

23. LAYERS OF THE RETINA-1
Image courtesy pcli.com

24. LAYERS OF THE RETINA-2
Image courtesy retinal physician

25. PRL