Evaluating the impact of artificial intelligence tools on the detection of chest injuries from medical imaging: A systematic review and meta-analysis.
Authors
Affiliations (1)
Affiliations (1)
- From the Department of Bioengineering (J.T.F.L., K.-J.C., I.G., S.M., S.H.), Imperial College London; Major Trauma Centre (J.T.F.L., D.F., S.H.), St Mary's Hospital, Imperial College Healthcare NHS Trust; and Traum@IC Research Group (J.T.F.L., S.H.), Centre for Injury Studies, and Statistical Advisory Service (J.E.), Imperial College London, London, United Kingdom.
Abstract
There has been a growing interest in the clinical application of artificial intelligence (AI) tools in medical imaging to aid diagnosis. This study conducts a systematic review of existing literature and performs a meta-analysis to compare the diagnostic performance of unassisted clinicians (CU) with clinicians assisted with AI (CA) in detecting traumatic chest injuries on diagnostic imaging. This systematic review was registered on the international Prospective Register of Systematic Reviews (CRD42024568478). A literature search was conducted on Ovid Medline, Ovid Embase, and the IEEE Xplore digital library, which included all studies evaluating the diagnostic performance of AI compared with a clinician for the detection of traumatic chest injuries on imaging in adults. The risk of bias was assessed using the quality assessment tool for diagnostic accuracy studies (QUADAS-2). Comparison between CA and CU groups was performed using meta-analysis for the primary outcome of diagnostic sensitivity and diagnostic time (DT) as a secondary outcome, with mean difference used as the effect measure. The search strategy identified 6,013 records. Following a full-text review, 20 studies were included, with 12 suitable for meta-analysis for rib fracture detection. The use of AI was associated with an improvement in sensitivity (CA, 0.88; CU, 0.76; mean difference, 0.12) and a reduction in DT (DT CA, 115 seconds; DT CU, 214 seconds; mean difference, -99 seconds). Artificial intelligence assistance can improve the diagnostic performance of clinicians. Clinicians assisted with AI were associated with an increase in the diagnostic sensitivity with a reduction in the DT to detect rib fractures on clinical imaging compared with CU. However, the overall quality of the evidence is poor, and further research into clinically useful models is required. Systematic Review and Meta-analysis; Level IV.