Deep Learning Based Identification and Classification of Diabetic Retinopathy

Abstract

A major factor contributing to blindness in diabetic people is diabetic retinopathy (DR). Early detection and classification are crucial for preventing vision loss. This study investigates the efficacy of deep learning models for identifying and classifying DR in retinal images. We deployed 9 different pre-trained models and one developed CNN model. The pre-trained models include ResNet 18,ResNet 50, DenseNet 121,  DenseNet  169,, DenseNet 201, EfficientNet  B5, MobileNetV2, InceptionV3, and Xception and CNN model was developed with  64,32 dense layers for binary classification differentiating DR and those without (non-DR). The models are trained and validated, evaluated on a retinal image dataset labelled for DR presence. The study analyses the accuracy of both the training, validation, and test datasets for each model in identifying DR. Notably, MobileNetV2 achieved the highest accuracy, outperforming the remaining models with an accuracy of 98 percent on test dataset.