Segmentation of the Optic Cup

The U-Net model for semantic segmentation and YOLO for object detection are the two most known artificial neural network-based methods. The YOLO model has been improved until today and presented under different version names. In this study, optic cup (OC) segmentation was performed on the REFUGE dataset by the YOLOv8, YOLOv9 models and U-Net, and then the results obtained by the models for the test images were evaluated. The U-Net, YOLOv8, and YOLOv9 models that we applied in this study achieved higher success for the OC segmentation. On average, for the REFUGE dataset test images, the U-Net model achieved 81% sensitivity, 82% Dice, and 70% Jaccard coefficient values, while the YOLOv8 and YOLOv9 models achieved 89% sensitivity, 81% Dice, and 69% Jaccard index values. This study presents an evaluation of the YOLOv8, YOLOv9, and U-Net models for the segmentation task. Furthermore, for the YOLOv8 and YOLOv9 models, which are mostly used for object detection, this study provides an example of the usability of the YOLOv8 and YOLOv9 models for segmentation purposes as well.