AI for Trauma CT

Computed tomography is the definitive imaging study in major trauma, and interpreting it quickly and correctly is critical for time-sensitive decisions such as angioembolization or surgery.

We have built deep learning systems for traumatic abdominal solid organ injuries, including detection and localization of liver, spleen, and kidney injuries on CT (Journal of Imaging Informatics in Medicine, 2024), a three-dimensional weakly supervised algorithm for splenic injury detection and sequential localization (International Journal of Surgery, 2023), and kidney trauma detection with external validation (International Journal of Surgery, 2025). This work received the National Innovation Award in 2025.

Our current NSTC-funded project develops an AI-based computer-aided diagnosis system for emergency whole-body CT in trauma: a unified model designed to simultaneously flag life-threatening injuries — intracranial hemorrhage, cervical spine fractures, aortic injury, hemopneumothorax, solid organ injuries, pelvic fractures, and retroperitoneal hemorrhage — as a clinical safety net embedded in the PACS workflow. The project builds on the Linkou CGMH trauma registry (>33,000 patients) and open datasets, with external validation partners across Chang Gung branch hospitals and international collaborators in trauma radiology AI.