Multistage Classification of Diabetic Retinopathy in Fundus Images using Hybrid Capsule Networks

Diabetic Retinopathy (DR) can occur in patients who have endured long-term diabetes. Treatment and diagnosis delays may lead to visual impairment. The main cause of diabetic patients’ DR is hyperglycemia. The blood vessels in the retina are affected by this. Manual diagnosis of diabetic retinal disease (DR) can be difficult because the disorder can modify the retina and result in structural alterations such as Microaneurysms (MAs), Hemorrhages (HMs), Exudates (EXs), and further blood vessel development. This research demonstrates utilizing a hybrid Deep Learning (DL) method to automatically identify and categorize DR in fundus images of the eye. Pre-processing is the initial phase. To achieve picture pre-processing, two processes are involved: (a) normalization and (b) gradient-based image contrast enhancement. Next, a U-Net Haar-like network is used to segment the preprocessed pictures. Ultimately, a novel hybrid classifier, termed Cap-MobileNet, combining Capsule Network and MobileNet V2, is proposed for multi-stage categorization of diabetic retinopathy. The proposed approach locates the afflicted lesions on the retinal surface and classifies DR images into five stages: no DR, mild DR, moderate DR, severe DR, and proliferative DR. Two datasets, APTOS and EyePACS, were used to test the proposed approach. The effectiveness of the system is assessed using a variety of measures, and the findings are contrasted with a few well-liked, cutting-edge, and recently developed models. The recommended approach greatly enhances the performance of fundus image DR detection. The prescribed improved strategy for multi-stage classification produced the best results, with 99.48% accuracy on the APTOS dataset and 99.16% accuracy on the EyePACS dataset. The proposed AI classification methodology holds the potential to provide a mass-screening platform for clinical deployment and telemedicine, as it is relevant and easily expandable to different ocular illnesses.