An Enhanced Deep Learning Approach for Detection and Classification of Retinal Diseases using OCT Images

Abstract

Ophthalmologists can make diagnoses based on the layers of the retina using optical coherence tomography (OCT), a non-invasive procedure produces cross section images of retina layer of the eye. As a result, it is a crucial modality for the identification and measurement of retinal disorders and illnesses. Ophthalmologists must spend a lot of time analysing the pictures since OCT produces many images for each patient. The OCT pictures of patients are divided into four groups in this work using deep learning models: Normal, Drusen, Diabetic Macular Edoema (DME), and Choroidal neo-vascularization (CNV). There are two distinct models suggested. Normal, Drusen, Diabetic macular edoema (DME), and Choroidal neo-vascularization are the four groups that may be classified utilising segmentation using curvature-based ROI and classification using the Folded RESNET101 algorithm (CNV). The proposed approach has the highest accuracy, sensitivity, and specificity of 0.99. The accuracy of binary classifier for the Normal is 0.99. These outcomes demonstrate their capability to function as a primary approach for ophthalmologists.