AUTOMATIC DETECTION OF DIABETIC MACULAR EDEMA FROM B-SCAN OCT IMAGES BASED ON PATTERN CLASSIFICATION TECHNIQUES

Abstract

Objective- This paper presents an automatic detection method to find the changes in
thickness of the layers of retina from Optical Coherence Tomography (OCT) B-Scan images for
resulting detection of Diabetic Macular Edema (DME) in diabetic patients.
Methods-In this study, using image processing methods initially the images are cropped according
to the region of interest, followed by conversion from RGB to gray scale, and then median filtered
method is used to de-noise them. Then, the six retinal layers are segmented using Graph-Cut Search
method. Region of interest is performed on the Optical Coherence Tomography (OCT) scan and the
automated retinal layers thickness measurements between the every two layers of the macula in
the regions are determined. Area enclosed between the every two layers is also estimated. Local
Binary Pattern (LBP) local texture features are extracted from segmented B-SCAN OCT images that
are subsequently trained and tested by Support Vector Machine (SVM) and Cascade Neural
Network(CNN); which are two different machine learning classifiers to classify whether the OCT
image is normal or DME affected. In this study, 55 datasets of OCT images (25 normal and 30 DME)
were used for classification.
Results- The performance results of both the classifiers are compared with respect to sensitivity,
specificity, accuracy, precision and F-Score. On comparison of these two classifiersthe CNN classifier
was better learns faster, highest accuracy than of SVM classifier, it has asensitivity of 95%,
specificity of 100%, accuracy of 96% ,precision of 100% and F-Score of 97%.
Conclusions- Thus, this algorithm can be used by ophthalmologists for early detection of Macular
Edema