Human–AI Collaboration & Clinical Deployment

An accurate model is not the same as a useful one. A central theme of our lab is understanding how AI assistance changes clinician behavior and performance — and how to integrate algorithms into clinical workflows safely.

We developed and validated a human-algorithm integration system for hip fracture detection in plain radiography (JMIR Medical Informatics, 2020), and we have compared different levels of AI support in the clinical interpretation of traumatic pelvic radiographs (npj Digital Medicine, 2025), asking when AI helps, when it biases, and how much autonomy an assistant should have.

Together with international colleagues, we have also mapped the trends, gaps, and future directions of AI for abdominopelvic trauma imaging (Abdominal Radiology, 2025).