Classification of Diabetic Retinopathy from OCT Images using Deep Convolutional Neural Network with BiLSTM and SVM

Abstract

Diabetic retinopathy (DR) is an eye disease that, if not diagnosed at an early stage, can cause vision loss and blindness in diabetic patients. Therefore, in the field of ophthalmology, optical coherence tomography (OCT) imaging is widely used in the early diagnosis and treatment of DR. However, it is arduous and time consuming to manually classify and detect DR from retinal OCT image. In this context, this paper proposed a deep convolutional neural network combined with Bidirectional long-short term memory and support vector machine (CNN-BiLSTM+SVM) for automatic DR classification from OCT image. Here, CNN-BiLSTM architecture is used as feature extractor where CNN extracts the local features and BiLSTM learns correlation among the extracted features. SVM is subsequently trained to diagnose diabetic retinopathy using the extracted features. The effectiveness of the proposed model is assessed on blind test dataset of 1000 images (250 per class) labeled into four classes. The proposed model has attained satisfactory results in the classification of diabetic retinopathy, with an accuracy of 99.30%, which also proves its competitiveness in comparison to other existing state-of-the-art.