A YOLO-based Method for Detecting Bone-Irregular in Chest X-rays

Bone-Irregular structures refer to small gaps in the bone cortex, leading to an uneven bone surface. Irregular bone structures require diagnosis through X-ray examination, and in severe cases, surgical treatment may be necessary. To address the issues of limited effective information, blurred images, and a small proportion of lesion objects in existing object detection methods for irregular bone structure detection, this paper proposes a YOLO-based detection method named YOLOBI. The method aims to assist doctors in identifying irregular bone structures in chest X-rays and reduce the probability of misdiagnosis. The core of this method includes: (1) employing a fusion strategy of multi-scale key features and local features to resolve the problem of insufficient effective information in irregular bone structure detection; (2) leveraging an attention mechanism to dynamically adjust the output of the neck network, improving the model's ability to distinguish irregular bone structures from background images and thus enhancing detection accuracy. Additionally, this paper develops a proprietary, non-public dataset, BoneIrr, for irregular bone structure detection research, used to validate the performance of irregular bone structure detection methods. The experimental results show that compared with the classical algorithm YOLOv8, YOLOBI improves the metrics mAP50 and mAP50-95 by 8.9% and 8.4% respectively.