Diabetic Retinopathy Detection Using Deep Learning with Optimized Feature Selection

Abstract

According to the International Diabetes Federation, there are 463 million diabetics worldwide. Due to alterations in lifestyle, the disease has had a significant negative influence on the quality of life for many people and is now seen as a global threat. Diabetic retinopathy (DR) is a leading cause of blindness among diabetic patients, emphasizing the need for early detection and intervention to prevent irreversible vision loss. It is caused by uncontrolled blood sugar and results in damage to blood vessels in the retina. Uncontrolled Type-1 diabetes and Type-2 diabetes both contribute to diabetic retinopathy. People with diabetes are more vulnerable to the severe effects of COVID-19. To prevent adverse consequences, early detection of the disease is essential. Therefore, automatic detection of diabetic retinopathy diseases is performed only by using computational techniques is a great solution. In this study, a deep learning model with an enhanced feature selection method has been constructed for the goal of early disease diagnosis. Also, the performance of the optimized dataset is evaluated using different machine learning methods such as Random Forest, FURIA, and decision tree. The proposed deep learning model achieved the highest accuracy of 93.5% with an optimized feature subset whereas random forest achieved a maximum accuracy of 92.26% with the optimized dataset and 91.2% and 89.4% with CFS-PSO and Information Gain.