Investigating the Clinical Value in Relation to Implementation and Use of an AI-Generated Fracture Algorithm Tool to Support Clinical Decision-Making

Abstract

Background/Objectives: The use of artificial intelligence (AI) in imaging departments is increasing in Europe. This study assesses the clinical value of an AI fracture algorithm by assessing ease of use, clinicians’ trust, and perceived barriers and benefits of this decision support tool in daily practice across two emergency departments (EDs) in Denmark. Methods: An online survey was distributed over four weeks (June–July 2025) to healthcare professionals interpreting radiographs in the ED at Lillebaelt Hospital. The survey included open-ended, closed-ended, and free-text questions addressing AI use. Additionally, an observational study was conducted, including workflow observations and time tracking of patient progression through the ED. Historical injury conference records from February 2023 to 2025 were reviewed to assess changes in patient management before and after AI implementation. Results: A total of 56 responses were obtained (24 male, 32 female). Most respondents reported a positive attitude toward the algorithm. Ease of use was rated satisfactory by 51 out of 56 participants, and 48 were satisfied with AI as a clinical decision support tool. Overall trust was high, with more than two thirds (n = 38) “agreeing” or “strongly agreeing” that the algorithm reliably detects fractures. However, an asymmetry in clinical trust was observed, whereby clinicians expressed greater confidence in their own assessments when the algorithm indicated the presence of a fracture than when it did not. Value stream analyses showed a delay of 6–23 min between radiograph acquisition and availability of the AI report. No differences were observed in the number of patients with treatment changes before, during, or after full implementation of the algorithm. Conclusions: In our limited study population, the AI fracture detection tool was overall well received by clinicians, although some observations indicate that implementation and workflow integration still require improvement. Larger studies are needed to validate the reported barriers and benefits of the AI fracture detection tool.

Ähnliche Arbeiten