Experience with AI-assisted Fracture Detection at a Pediatric Hospital.
Authors
Affiliations (2)
Affiliations (2)
- Texas Children's Hospital, Edward B. Singleton Department of Radiology, Texas, United States, Houston.
- Baylor College of Medicine, Department of Radiology, Texas, United States, Houston.
Abstract
Artificial intelligence (AI) is increasingly used in radiology. For pediatric patients, the only commercially available computer-aided detection/diagnosis tool currently available is for fracture detection. Here, our experience at a pediatric hospital with a 2-month trial of a commercial AI fracture-detection tool is described. Although the tool was sometimes helpful, particularly for subtle fractures and as confirmatory support in borderline cases, its overall value was limited by cost, false-positive and false-negative results, workflow inconsistency, and the burden of post-deployment monitoring. In our experience, aggregate performance alone did not capture the practical issues most relevant to clinical adoption, including delayed or missing AI results, provider confusion when AI and radiologist interpretations differed, and the infrastructure required for sustainable oversight. Our experience suggests that fracture-detection AI should be evaluated not only by diagnostic performance, but also by real-world clinical utility, workflow reliability, and implementation burden and provides real-word insight into these issues.