AI model distinguishes lateral malleolar avulsion fractures from subfibular ossicles on ankle radiographs

A two-stage deep learning model (RetinaNet + MobileNetV2) achieved AUC 0.887 on external test set for differentiating lateral malleolar avulsion fractures from subfibular ossicles on ankle X-rays, outperforming radiologists and improving their accuracy when used as an aid.

• Multicenter retrospective study of 2,121 patients across two centers, with an external test set.
• Radiologists' diagnostic accuracy improved significantly when assisted by AI predictions and saliency maps (exact figure not reported in source).
• Retrospective design; external validation limited to a single test set; no prospective clinical implementation data.

RadPigeon summaries are original and for information only. They are not clinical advice.