All evidence
Evaluation of the use of artificial intelligence in the detection of appendicular skeletal fractures in adult patients consulting in an emergency department
Data
Consecutive data of limb x-rays from the emergency department over 1 year
4475 patients in total
Design
Retrospective study
Only examining radiographs with discrepancies between the AI results and the ED physician’s diagnosis
Emergency physicians re-evaluated the discrepant radiographs with AI assistance
Ground truth:
Concordance between AI and report
In the event of discrepancies, adjudication by a radiologist
Results
603 discordant cases
282 = confirmed fractures
220 were not identified by the emergency physician and 62 were correctly identified
Proportion of fractures missed by the ED physician without AI VS Proportion of fractures missed by the ED physician with AI
78.0% VS 45.3%
Diagnostic error rate decreased by 41.9% with AI assistance
