Detection of Diabetic Retinopathy using Convolutional Neural Network

Abstract

A medical disorder called diabetic retinopathy arises when diabetes mellitus damages the retina. It takes a lot of time to diagnose diabetic retinopathy using coloured fundus pictures since skilled doctors are needed to recognise the existence and importance of numerous tiny characteristics. In this study, we suggest a CNN-based method for identifying diabetic retinopathy in fundus photographs. A novel segmentation method utilising Gabor filters is employed to prepare the data for the model's training. Data augmentation is used to increase the dataset since it is insufficient to train the model. Our segmentation model can identify the presence of DR and recognise complex characteristics in fundus pictures. The model is effectively trained using a top-tier graphics processing unit (GPU). Impressive outcomes are shown using the publicly accessible Kaggle Dataset, especially for a challenging categorization assignment. Our suggested CNN obtains a specificity of 94% on the training dataset of 14,650 pictures.