AUTOMATIC DETECTION OF MENISCUS TEARS FROM KNEE MRI IMAGES USING DEEP LEARNING: YOLO V8, V9, AND V10 SERIES

OpenAlex İngilizce

Meniscal tears are a disease that occurs in the knee joint and negatively affects people's mobility. In this study, the performance of the state-of-the-art (SOTA) YOLO (You Only Look Once) models, in particular YOLOv8l, YOLOv8x, YOLOv9c, YOLOv9e, YOLOv10l, and YOLOv10x, for the detection of meniscal tears was investigated. The algorithms were trained and tested with data from magnetic resonance imaging (MRI). In our study, the YOLOv9e model showed the highest performance and achieved the best results in the training phase with a mAP50 of 0.91807, a precision of 0.87684, a recall of 0.93871 and an F1 score of 0.90672. This study makes a unique contribution to the field with its advanced algorithms and comprehensive performance analysis. The findings show that deep learning algorithms are suitable for clinical use in the automatic detection and localization of meniscal tears. In this way, the possibility of early diagnosis increases, and patients can be directed to the right treatment, preventing joint problems that may occur in the future. In future studies, it is aimed to increase the generalization capabilities of the models with larger data sets and different anatomical structures.