Automatic Rib Fracture Detection and Localization from Frontal and Oblique Chest X-Rays

Several deep learning-based object detection techniques in medical imaging have been proposed. Chest X-rays are widely used for detecting thorax diseases due to the convenience and low radiation dose compared to Computed Tomography (CT). However, the research on rib fracture detection in chest X-rays is still inadequate. Most of the research primarily focused on frontal CXR and some on lateral CXR. No study of rib fracture detection on oblique view CXR has been previously proposed. Due to the overlapping characteristic of human ribs, the oblique view can help radiologists to recognize the fractured ribs that are blocked in the frontal view. In this paper, we employed a YOLOv5 model along with the techniques of data augmentation and image enhancement for rib fracture detection. We trained and evaluated on E-DA dataset, a private dataset collected from E-DA Cancer Hospital containing frontal and oblique chest X-rays. The developed model can detect fractured ribs in both projection views of CXR.