Wrist Fracture Detection in X-Ray Images with YOLO Algorithms
Ayşe Aybilge MURAT Mustafa Servet KIRAN
AbstractSignificant advancements have been made in the analysis of medical images through systems developed using deep learning methods. The detection of fractures using computer vision is one of the current areas under investigation. Identifying bone fractures in X-ray images is a time-consuming process that requires specialized expertise. Pediatric bone fractures, in particular, are common and situations that require prompt treatment to prevent future complications. Consequently, deep learning-based detection systems hold great importance in supporting clinical decision-making processes and saving time for specialists. This study evaluates five recent YOLO algorithm versions (YOLOv8-YOLOv12) for pediatric wrist fracture detection in X-rays. The Pediatric Wrist Trauma X-ray dataset (GRAZPEDWRI-DX) was used, and training was conducted on 10,300 X-ray images. The models' accuracy, speed, and generalization capabilities were analyzed, and it was observed that the YOLOv9s and YOLOv12m models achieved the best performance (0.944 mAP50 and 0.90 Recall), with all trained models showing similar results. This study aims to demonstrate the performance of YOLO-based models in the automatic detection of bone fractures in X-ray images, contributing to the acceleration of the diagnostic process for these common injuries.