Diagnosis of Retinal Diseases by Classifying Lesions in Retinal Layers using a Modified ResNet Architecture

Abstract

Optical Coherence Tomography (OCT) was first introduced in the 1990’. It utilizes the concept of interferometry to create a cross-sectional map of the retina., accurate within 10–15 microns. Identifying the actual diseases occurring in retina layer., is a challenging task. There exist several automated techniques for disease classification like image processing., deep learning. Unfortunately., these techniques often produce error., lower precision., excessive memory localization., inefficiency in computation., further interpretation of human experts. In this paper., we have proposed a method for automatic classification of 3 categories of retinal diseases that include diabetic macular edema., Drusen., Choroidal Neovascularization. A modified ResNet architecture with transfer learning framework is used to make better feature extraction for small patches. This modification includes adding three new layers which are Convolution layer., Batch Normalization and Activation function relu layers. Modification is added at the end of convolution layers in a pretrained Resnet framework. These layers are inserted in the ResNet50 architecture for accurate discrimination and robust feature extraction of OCT images with better efficiency than the traditional networks. Experimental results demonstrate that our method obtained accuracy value 99.81%. Our proposed model provides reliable classification for small lesions., helpful in clinical diagnostic to provide user-friendly eye check-ups.