Artificial intelligence-assisted training for rib fracture interpretation: a prospective study in undergraduate medical students.
Authors
Affiliations (6)
Affiliations (6)
- Department of Trauma and Emergency Surgery, Department of Surgery, Chang Gung Memorial Hospital, Linkou, Taoyuan City, Taiwan.
- Chang Gung University, Taoyuan City, Taiwan.
- Department of Biomedical Engineering, National Taiwan University, Taipei, Taiwan.
- Chang Gung Memorial Hospital Surgical Training Academy and Research Center, Taoyuan City, Taiwan.
- Department of Trauma and Emergency Surgery, Department of Surgery, Chang Gung Memorial Hospital, Linkou, Taoyuan City, Taiwan. [email protected].
- Chang Gung University, Taoyuan City, Taiwan. [email protected].
Abstract
Chest X-rays (CXRs) are essential in trauma care but have limited sensitivity for rib fracture detection, leading to frequent missed diagnoses. Artificial intelligence (AI) has shown potential to improve diagnostic accuracy, yet its role in radiology education remains underexplored. This study evaluated the impact of AI-assisted training on early diagnostic performance and confidence in rib fracture detection on trauma CXRs. In this prospective observational study, 26 undergraduate medical students (UGY) completed three sequential sessions: baseline unassisted interpretation of 50 CXRs (Session 1, S1), AI-assisted interpretation of the same cases (Session 2, S2), and interpretation of 50 new CXRs without AI assistance (Session 3, S3). Diagnostic performances and confidence levels were compared across sessions. AI-assistance (S2) significantly improved all performance metrics, with increases of 26.7% in accuracy, 41.0% in sensitivity, 25.1% in specificity, 35.6% in F1 score, and 31.4% in precision (all pā<ā0.01). Performance in S3 declined compared to S2 but remained higher than baseline for accuracy (+ā13.3%, pā=ā0.010) and precision (+ā13.7%, pā=ā0.010) compared to baseline. Confidence levels showed sustained improvement across all sessions (pā<ā0.001). Agreement analysis in AI-misclassified cases suggested possible automation bias in S2 and carryover effects in S3. AI-assisted training significantly enhances early diagnostic performance and confidence in rib fracture detection on chest radiographs, a key competency in trauma and emergency care, with partial skill retention after AI withdrawal. Integrating AI into early trauma imaging training may strengthen radiology training but requires strategies to mitigate automation bias and foster independent judgment.