Artificial intelligence in emergency skeletal X-ray: post-deployment monitoring and clinical impact of incorrect AI results.

OBJECTIVES: This study evaluates the impact of the artificial intelligence (AI) application BoneViewTM within a radiographer-supervised clinical workflow and subsequent effects on patient management. We examine false AI findings to understand the clinical consequences in skeletal X-ray examinations of suspected fracture injuries. MATERIALS & METHODS: This retrospective study was conducted at Bærum Hospital, Norway, where BoneViewTM was introduced within an AI-assisted radiographer-supervised emergency skeletal X-ray workflow in September 2023. The study included patients who were referred to emergency skeletal X-ray in January 2024. We calculated diagnostic accuracy of AI compared with AI-assisted radiologists, and we examined patient management following the introduction of the new AI-assisted workflow. RESULTS: The study included 1248 patients. AI alone demonstrated an overall sensitivity of 0.95 (95% CI 0.92 to 0.97) and specificity of 0.90 (95% CI 0.88 to 0.92) when compared to AI-assisted radiologists. Compared to AI's initial suggestion, radiographer supervision altered the subsequent patient pathway in 20% of the cases, with the likelihood overriding AI being higher when AI results were later found to be incorrect (i.e. false positives or false negatives). Notably, the AI-assisted workflow only led to premature discharge of one patient who required treatment, and this patient was recalled to hospital for treatment following the radiologist's report. CONCLUSION: AI demonstrates strong diagnostic accuracy with high sensitivity and specificity in emergency skeletal X-ray examinations. Radiographers may play a crucial role in mitigating erroneous clinical decisions due to false AI-results and ensure appropriate patient management.