YOLOv8-Based Real-Time Bone Fracture Detection and Localization Using Deep Learning

Abstract

Bone fractures are one of the most common orthopedic injuries and require fast and accurate diagnosis for effective treatment and patient recovery. This thesis presents an intelligent web-based bone fracture detection system using YOLOv8 and Deep Learning techniques for automated analysis of X-ray images. The YOLOv8 object detection model is trained using a dataset containing 3,316 training images and 399 validation images covering seven fracture classes including Elbow Positive, Fingers Positive, Forearm Fracture, Humerus, Humerus Fracture, Shoulder Fracture, and Wrist Positive. The system performs real-time fracture detection and localization by identifying fracture regions through bounding box prediction, thereby assisting medical professionals in faster and more accurate diagnosis. Experimental results prove that the proposed model reaches a mean Average Precision (mAP50) of 86%, highlighting the efficiency of the system in handling complex medical imaging tasks with high consistency and productivity. The proposed framework provides a cost-effective and scalable solution for initial fracture screening and has important potential for future integration into clinical healthcare systems and intelligent medical diagnostic applications.

Keywords