Diabetic retinopathy classification using improved metaheuristics with deep residual network on fundus imaging

Diabetic Retinopathy (D.R.) is a common illness occurring among persons with diabetes. Automated systems utilizing deep learning (DL) hold the potential for improving the accuracy and effectualness of D.R. screening, which is significant for early recognition and treatment to avert vision loss. This article focuses on the design of Improved Metaheuristics with DL-based D.R. Grading on Retinal Fundus Images (IMDLDRG-RFI) technique. The IMDLDRG-RFI technique primarily employs an adaptive median filtering (A.M.F.) approach for noise removal. Besides, the IMDLDRG-RFI technique uses the ResNet101 model as a feature extractor with an improved salp swarm algorithm (I.S.S.A.) as a hyperparameter optimizer. Finally, a crisscross optimization algorithm (C.S.O.) performs the D.R. classification process with a Wide and Deep Fourier Neural Network (W.D.F.N.N.) model. The experimental result analysis of the IMDLDRG-RFI technique is tested on a benchmark D.R. dataset. The results show that the proposed approach can accurately classify D.R., with an \(acc{u}_{y}\) $acc{u}_y$ of 99.90%, \(sen{s}_{y}\) $sen{s}_y$ of 99.55%, and \(spe{c}_{y}\) $spe{c}_y$ of 99.91%. This designates that the proposed approach can potentially be a valuable tool for the automated detection and classification of D.R.