Mild DR Detection from CLAHE Fundus Images using Experimental Minimal CNN Model with Batch Normalization

Abstract

Vision impairment is one of the major problems affecting middle-aged individuals due to uncontrolled blood sugar levels, commonly known as Diabetic Retinopathy (DR). The small abnormalities in the retinal capillaries, called microaneurysms and intra retinal bleeding, are the initial symptoms of Diabetic Retinopathy. Clinical identification of Diabetic Retinopathy is a time consuming and difficult process due to limitations in resources and experienced doctors. Early detection is crucial in avoiding the progression of Diabetic Retinopathy, highlighting the importance of an automated DR detection method to identify symptoms in its early stages. In this paper, researchers developed an Enhanced Minimal Convolutional Neural Network (EMCNN) model to classify Mild-DR and No-DR fundus images using a binary classification process. The fundus images were pre-processed using Contrast Limited Adaptive Histogram Equalization (CLAHE) method before passed through the network. EMCNN is an experimental model that enjoys a minimum number of layers and batch normalization to minimize the training effort. Finally, the EMCNN model is compared to existing models in terms of accuracy and efficiency