A Customized Deep Learning Algorithm for Prediction of Eye Diseases from Color Fundus Photography

Abstract

In the recent years, considerably most of the people suffer from severe eye related diseases due to irregular check-up and high consuming time. The main of the work is to recognize to major different kind of eye related diseases such as Cotton-wool spots, Fibrosis, Fundus neoplasm, Maculopathy, Myelinated nerve fiber, Optic atrophy, Peripheral retinal degeneration and break, Possible glaucoma, Preretinal hemorrhage, Severe hypertensive retinopathy through Convolution Neural Network and detect diseases in less time. Retinal fundal images are collected from kaggle source and preprocessed by performing gray scale conversion, image enhancement, histogram equalization and standardization techniques. By comparing the existing architecture such as mobile net, Resnet50 and VGG19 with the customized new architecture and show better performance than the existing one by comparing its quantitative analysis and the result is obtained by predicting accurate diseases with less training and validation time with high accuracy.