Juxtaposing Deep Learning Models Efficacy for Ocular Disorder Detection of Diabetic Retinopathy for Ophthalmoscopy

Abstract

The reticence of ocular disorder in the community has been a major deterrent to one being visually impaired before getting diagnosed. The genesis lies deeper in the inefficiency of ophthalmologists configuring the root cause of defacement. One such predicament is Diabetic Retinopathy (caused due to changes in retinal blood vessels) which is in an upsurge causing apprehension for vision loss. Meagre research has been carried out using different learning techniques giving a vague prospect of a cogent solution. In this paper, we have subsumed potent Deep Learning (8-Layer CNN) and Transfer Learning architectures (MobilenetV2, DenseNet121, InceptionV3, ResNet50, VGG16) for deducing the potentiality of a person having Diabetic Retinopathy using a 02-Class model with collating varied dataset namely APTOS 2019 and HRF Image Database begetting preeminent accuracy results with metrics comprehended such as f1 Score, Area Under Curve, Cohen’s Kappa Score for corroboration.