Multi-Level severity classification for diabetic retinopathy based on hybrid optimization enabled deep learning

Abstract

The disease with the fastest global growth rate, diabetes, has numerous complications. Diabetic retinopathy (DR), which results in retinal lesions that affect vision, is one of these conditions. There are many DR levels, from moderate to extreme. Early DR detection and treatment can lower the risk of losing one's vision. It takes a lot of time and effort to identify DRs today since competent ophthalmologists must look at digitised, coloured retinal fundus images. This study uses SqueezeNet and Deep Convolutional Neural Network (DCNN) in two rounds to provide a novel multilevel severity classification of DR. The fundus image is initially categorised using Squeezenet into the normal or abnormal class of DR. Here, the Fractional War Strategy Optimization (FrWSO), which combines War Strategy Optimization (WSO) and Fractional Calculus (FC), tunes the SqueezeNet. The severity level for the abnormal images is determined using DCNN in the second level decomposition. Additionally, the proposed Fractional War Royale Optimization (FrWRO) algorithm, which combines Battle Royale Optimization (BRO) with FrWSO, adjusts the weight of the DCNN. The tests show that the accuracy, sensitivity, and specificity of the suggested approach are 0.911, 0.898, and 0.913, respectively, yielding the best performance.