Diabetic Retinopathy Classification Using Deep Residual Network with Remora Tuna Swarm Optimization

Diabetic retinopathy (DR) is a harmful eye state, which influences diabetic patients. Unless earlier detected, it affects the retinal portion and ultimately causes eternal vision loss. An earlier diagnosis of DR is more vital for treatment purposes. Though, DR diagnosing is a highly complicated process, which needs a knowledgeable ophthalmologist. In this work, Deep Residual Network-Remora Tuna Swarm Optimization (DRN-RTSO) is introduced for DR classification. A fundus image considered is pre-processed utilizing an adaptive wiener filter. Then, lesions are segmented employing SegNet includes Microaneurysms, Haemorrhages, Soft Exudates and Hard Exudates. Thereafter, blood vessel segmentation is conducted on the pre-processed image using Dense U-Net. Afterwards, feature extraction is carried out considering input fundus image and segmented outputs. At last, DR is classified into normal, proliferative, mild non-proliferative (NPDR), severe NPDR and moderate NPDR utilizing DRN that is tuned utilizing RTSO. The RTSO is devised by incorporating Remora Optimization Algorithm (ROA) with Tuna Swarm Optimization (TSO). In addition, DRN-RTSO attained maximal accuracy of 91.5%, negative predictive value (NPV) of 93.3%, positive predictive value (PPV) of 91.1%, true negative rate (TNR) of 91.7% and true positive rate (TPR) of 92.3%.