Experimental Demonstration of Retinal Image Analysis and Disease Detection using Deep Learning based Detection and Classification

Deep learning (DL) enhances the analysis of medical images in ophthalmology for the rapid identification of retinal disorders. This study introduces retinal image analysis and disease detection using the deep learning detection and classification (RIADDL) model, which is aimed at categorizing retinal conditions through OCT images. The dataset comprises 109,312 high-resolution OCT images. These images are divided into four categories: CNV, DME, DRUSEN, and NORMAL. To tackle challenges related to imbalanced datasets and low-contrast images, the research utilizes efficient data preprocessing and augmentation techniques. CLAHE enhances image contrast, while techniques like flipping, rotation, cropping, and scaling boost model robustness. Fuzzy k-means (FKM) clustering is utilized to precisely detect diseased regions. The RIADDL Model utilizes a simple Convolutional Neural Network (CNN)-based Sparse Residual Network (SRN) to lower computational complexity while ensuring high precision. This network substitutes conventional 5×5 convolutions with several 3×3 layers, reducing parameters and computational expenses. The RIADDL Model provides a reliable and accurate approach for identifying retinal disorders, demonstrating the potential for application in clinical environments to facilitate early diagnosis and avert vision impairment.