Improving YOLO26-OBB for Oriented Fracture Detection in X-Ray Images

In this study, we propose a novel architecture based on YOLO26-OBB for oriented fracture detection in X-ray images. While traditional approaches such as horizontal bounding boxes (HBB) and segmentation are widely used, they often fail to capture the true orientation and precise boundaries of fractures. Oriented bounding boxes (OBB) provide a more accurate representation by aligning with the fracture direction, improving localization quality. To enhance detection, we introduce lightweight architectural modules including a Rotational Feature Enhancement Block (RFEB) for capturing orientation-specific features and an Edge- Aware Attention Module (EAAM) for improving boundary refinement. Additionally, training strategies are applied to handle the limitations of small datasets like FracAtlas. Experimental results show improved mAP@0.5:0.95 and recall with consistent localization under stricter IoU thresholds and minimal impact on inference time.