A Technique for Retinal Detachment Detection Manipulating YOLOv8 Models

Abstract

A retinal detachment is a serious condition resulting in the retina detaching from its support layers, which are just beneath it. If untreated, this can cause blindness. To detect the classes of retinal detachment in various images, this paper introduces a novel method employing deep learning techniques involving the YOLOv8 algorithm. Notably, this marks the first use of the YOLOv8 model for retinal detachment detection. Retinal detachment can be identified with high precision using images obtained from Optical Coherence Tomography (OCT). The proposed work assesses the performance of these models using metrics such as mAP50, recall, and precision by training five YOLOv8 models: YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l, and YOLOv8x. Among these, the YOLOv8s model had the best performance with a mAP50 of 0.985, a recall of 0.97, and a precision of 0.968. The other models had the following mAP scores: YOLOv8n (0.949), YOLOv8m (0.906), YOLOv8l (0.889), and YOLOv8x (0.907). This demonstrates that the proposed system works effectively in detecting retinal detachment, resulting in highly accurate results mined from complex medical data sets and imaging, thereby making it an important tool in medicine.