Multiclass Identification and Classification of Diabetic Retinopathy using Transfer Learning: A Comprehensive Approach for Enhanced Diagnostic

Diabetic retinopathy, a commonly observed disorder associated with diabetes mellitus, is distinguished by the emergence of irregularities in the retina that adversely affect visual capabilities. In the absence of timely identification, the condition above has the potential to advance and result in compromised visual function. While proven to be effective, the process of manual diagnosis conducted by ophthalmologists is characterized by time-consuming procedures, labor-intensive efforts, significant costs, and a susceptibility to misdiagnosis. On the other hand, computer-aided diagnosis procedures provide a highly efficient and precise alternative. The utilization of deep learning techniques, namely in medical picture interpretation, has demonstrated improved effectiveness. This study introduces an innovative deep-learning methodology for detecting and classifying diabetic retinopathy automatically. This study proposes the utilization of a Convolutional Neural Network model based on ResNet50 architecture to identify and classify retinal pictures into five distinct categories, which correspond to different phases of diabetic retinopathy. The model demonstrates a validation accuracy of 95.01% and a training accuracy of 98.70%, accompanied by a Cohen Kappa score of 0.96 with negligible loss. The efficiency of the proposed model is assessed by utilizing the Diabetic Retinopathy Detection dataset, demonstrating its capability to automatically identifying diabetic retinopathy. This research has the potential to be applied within the healthcare domain, specifically to diagnose retinal complications in individuals with diabetes.