Automated detection and multi-stage classification of diabetic retinopathy through CNN

Diabetic retinopathy (DR) is a prominent cause of vision loss. It is one of the most significant sources of eye disease among individuals who have had diabetes for a long time. A good prognosis is dependent on early identification of the disease but determining the exact stage of diabetic retinopathy from the color fundus images is a significant challenge. The subtle distinction among different DR stages (Level 0 to Level 4) and the existence of many structures of varying shapes and sizes makes manual recognition challenging and consumes more time. Machine learning (ML) and deep learning (DL) based approaches are the best alternative for automatically detecting and analyzing DR advancement. The proposed work presents a comparative analysis of the machine learning and deep learning-based methods for both binary and multi-level classification of DR when tested on APTOS 2019 Blindness Detection Kaggle Dataset. Support Vector Machine (SVM) is more influential among the different ML classifiers. It is compared with a Convolutional Neural Network (CNN) designed from scratch, a pre-trained CNN-based model, VGG-16, and other existing methods to classify fundus images for binary (DR and No DR) and multiple levels (L0-L4) of DR based on the severity. Experimental results validate the effectiveness of VGG-16 for both binary and multi-level classification in terms of different evaluation matrices.