Enhanced Detection and Segmentation of Diabetic Retinopathy Lesions Using a Region of Interest Strategy

Diabetic retinopathy is a common microvascular complication of diabetes that can lead to vision loss. Experts diagnose it by identifying retinal lesions, including hard exudates, soft exudates, hemorrhages, and microaneurysms, the earliest indicators. These lesions often appear near the macula, making this region crucial for early diagnosis. This study proposes a YOLOv12based approach for detecting and segmenting retinal lesions, focusing on the macula. We trained two models: one for region extraction using the Indian Diabetic Retinopathy Dataset and another for lesion detection with the Dataset for Diabetic Retinopathy. The model achieved a mean Average Precision of 0.4530 on the validation set and 0.3020 on the testing set for lesion segmentation.