AI Model to Detect Bone Fracture

Abstract: Bone fractures represent one of the most prevalent categories of traumatic injury encountered in emergency clinical settings, and their prompt identification from radiographic images is paramount to ensuring effective patient management and reducing the risk of long-term musculoskeletal complications. Conventional manual examination of X- ray images is inherently subject to inter-observer variability, fatigue-induced diagnostic errors, and throughput constraints, particularly in high-volume accident and emergency departments. This paper presents a comprehensive survey and an original deep learning framework leveraging the You Only Look Once version 8 (YOLOv8) single-stage object detection architecture for real-time, automated detection and spatial localisation of bone fractures in plain radiographic X-ray images. The proposed pipeline integrates Contrast Limited Adaptive Histogram Equalisation (CLAHE)-based image enhancement, mosaic data augmentation, and a structured transfer learning strategy founded on MS COCO pre-trained weights to maximise generalisation across diverse fracture morphologies and anatomical regions. A consolidated dataset of 8,742 annotated radiographs spanning seven skeletal regions was employed for training, validation, and testing under stratified partitioning. Experimental evaluation demonstrates that the proposed YOLOv8m model achieves a mean Average Precision (mAP@0.5) of 91.4%, a clinical sensitivity of 92.7%, and a specificity of 89.3%, with a real-time inference throughput of 56 frames per second. Systematic comparative benchmarking against VGG-16, ResNet-50, Faster R-CNN, and YOLOv5s confirms the superiority of the proposed approach. An ablation study further validates the individual contributions of CLAHE pre-processing, mosaic augmentation, and transfer learning to overall detection performance. The findings establish YOLOv8 as a clinically viable, decision-support technology for automated fracture screening in radiology workflows.

Keywords: Bone Fracture Detection, YOLOv8, YOLO, Deep Learning, X-Ray Image Analysis, Medical Imaging, Object Detection, Convolutional Neural Networks, Computer-Aided Diagnosis, Transfer Learning, CLAHE