The effectiveness of automated classifiers for diabetic retinopathy in a methodical diagnostic program

It is estimated that 600 million people will have diabetes by 2040, with one-third developing diabetic retinopathy. Diabetic retinopathy (DR) is among the most debilitating complications of diabetes and can result in blindness. The grade level must be monitored so that the appropriate treatment option can be implemented in a timely manner. There is a need for efficient automated methods to detect diabetic retinopathy and categorize its severity stage. These methods would help alleviate the workload of ophthalmologists and minimize discrepancies in diagnosis among manual readers. This study employed an algorithm based on deep learning to discern the presence and stages of diabetic retinopathy by analyzing images relating to the color-coded retinal fundus. Various class configurations have been utilized in the categorization of severity stages of diabetic retinopathy. The suggested model was tested in this study using the Indian Diabetic Retinopathy Image Dataset from Kaggle into the following four classification levels; no_DR, mild, moderate, and severe. Using transfer learning RestNet-50 to conduct multiple trials with varying severity levels is intriguing. The greatest achievable categorization accuracy is 75.95%. In terms of performance, the ResNet-50 model beats the Inception-V3 and VGG models.