Diagnostic Accuracy, Time to Diagnosis Improved With AI Assistance

Objective: To compare the diagnostic accuracy of an artificial intelligence (AI)-assisted radiologist to one who is not AI assisted. Design: Systematic review and meta-analysis. Methods: Randomized and cohort studies assessing AI’s ability to improve clinician’s diagnostic accuracy were included. The focus was on studies comparing AI-assisted and AI-unassisted clinicians. Standard study quality and risk-of-bias assessments were performed per guidelines for systematic reviews. The primary outcome was the sensitivity to detect chest injuries on CT scan. Secondary outcomes included positive predictive value and diagnostic time. Subanalysis accounting for reader experience stratified by 5 years was also performed. Results: 20 studies met the inclusion criteria, with all but one assessing the accuracy of CT scan interpretations while one assessing the accuracy of plain chest radiographs. A significant number of studies had a high risk of bias. The gold standard assessment was more...