Long Bone Fracture Detection Using DEtection TRansformers

Abstract

Accurate diagnosis of long bone fractures is crucial to prevent complications such as delayed healing or permanent deformities. In medical imaging, machine learning (ML), particularly deep learning (DL) models, has emerged as a powerful tool for enhancing diagnostic accuracy, reducing human error, and streamlining workflows through automation. This study investigates the DEtection TRansformer (DETR), a transformer-based DL model, for automated fracture analysis in X-ray images. A dataset of 3,000 long bone radiographs, manually labeled and annotated by an expert, was preprocessed and augmented to support robust model training and evaluation. DETR achieved a mean Average Precision (mAP) of 80.9%, demonstrating strong sensitivity and reliable localization of fracture regions. Its end-to-end architecture and ability to capture global contextual dependencies make it particularly effective in identifying subtle or complex fracture patterns. These results highlight DETR’s potential as a clinically relevant tool, with future improvements anticipated through larger datasets, advanced augmentation strategies, and architectural refinements.