Severity Levels Categorization for Detecting Diabetic Retinopathy Using YOLOv7

A disease that damages the eyes called diabetic retinopathy (DR) is more frequently found in adults who are working. Normal, moderate, and proliferate are the three levels of diabetic retinopathy severity. The YOLOv7 algorithm and PyTorch framework are used in this paper to propose a novel model for assessing the DR severity in its various phases. For the precise and effective detection of lesions in retinal images, the suggested method makes use of a convolutional neural network model, data augmentation, and reinforcement learning techniques. The abnormality or damage visible in the retina of the eye is referred to as a lesion in retinal imaging. The validation set’s mean average precision for the model, which was trained on the blindness detection dataset, was 0.921 and 0.902 for intersection over union thresholds of 0.5 and 0.95, respectively. The proposed model gives promising outcomes for early diagnosis and treatment of DR, which is essential in preventing serious outcomes like blindness.