Diabetic Retinopathy Detection, Segmentation and Classification Using YOLO v8, MultiResUNet and ResNet 18

Diabetic retinopathy (DR) is a major global cause of vision impairment; accurate retinal lesion segmentation and early detection are essential for both diagnosis and treatment. In this work, a two-stage framework for automatic deep learning-based DR detection and segmentation is presented. The initial step involved training a YOLOv8-based object identification model to identify and categorize retinal abnormalities. In the second step, detailed lesion masks were created using a MultiResUNet segmentation model. In order to accurately localize retinal defects, the study first trains an object identification model (YOLOv8) to recognize and locate areas of fundus pictures impacted by DR. This stage guarantees accurate identification of crucial areas that need more investigation. The project’s next goal is to use MultiResUNet to precisely segment the lesions that were found. The model helps determine the severity of the disease by producing high-quality segmentation masks that improve viewing of the afflicted areas. The model is assessed using important performance measures including precision, recall, and segmentation accuracy in order to confirm the efficacy of the suggested strategy. This guarantees the system’s accuracy and dependability for practical uses.