Diabetic Retinopathy detection using Region based convolutional neural network (RCNN) and classification based on Retinal Fundus Images

Diabetes mellitus is a chronic disease in which the body either fails to produce sufficient insulin or cannot efficiently utilize it due to a hormonal imbalance. Individuals with diabetes are prone to visual impairments, such as diabetic retinopathy (DR) and glaucoma, as well as other complications, including neuropathy, nephropathy, and cardiomyopathy. Among these, DR is one of the most prevalent consequences of diabetes. Detecting morphological alterations in the retina is challenging, necessitating an effective automated detection system. Advances in deep learning models and image processing techniques have facilitated the early prediction of DR. Numerous research studies on DR diagnosis systems have been published in recent years, highlighting the growing significance of deep learning (DL) models in medical image classification. With this objective in mind, this study introduces the Region-based Convolutional Neural Network (RCNN) model to identify DR in retinal fundus images. The RCNN approach incorporates noise removal using Gabor filtering (GF) and a segmentation strategy that enhances region growth. Furthermore, the seed location and thresholds for the region-expanding segmentation method are improved using the Glowworm Swarm Optimization (GSO) system. In adding, SqueezeNet with a class attention learning layer is used for feature vector extraction. The RCNN-GSO methodology is validated using the benchmark MESSIDOR dataset, demonstrating superior performance over recent approaches, with a maximum accuracy of 98.22%. The simulation results confirm that the RCNN-GSO technique outperforms state-of-the-art approaches for DR diagnosis.