Written by Doug Wallace
Spoon Feed
This retrospective study assessed the benefit of AI-assisted chest x-ray interpretation for non-radiology residents when compared to radiology residents. Pneumothorax and pulmonary nodule detection, in particular, demonstrated meaningful gains in sensitivity and accuracy for non-radiology providers.
The AI “Arrow Sign”
Chest radiographs (CXR) are ubiquitous in medical practice, and a commonly used primary diagnostic in the ED setting. Many rural and resource limited settings lack 24-hour radiology coverage. Can AI improve CXR interpretation accuracy and sensitivity?
Over 500 radiographs were retrospectively assessed in this European study comparing AI benefits in CXR interpretation for non-radiology residents (NRRs) vs radiology residents (RRs). Four discrete domains were assessed: pleural effusions, pneumothoraces, consolidations, and pulmonary nodules. Generally, performance gains in terms of sensitivity and accuracy were minimal for RRs. Marked improvement was noted in all four domains for NRRs, with the most significant benefits in the detection of pulmonary nodules (53% sensitivity increase, 7% accuracy increase) and pneumothoraces (30% sensitivity increase, 2% accuracy increase).
The authors posit AI technology could serve as a “second reader” if radiology coverage is lacking but recommend caution in blindly trusting the AI.
How will this change my practice?
It’s best to keep in mind this study is not EM specific, and none of the residents involved were categorical EM. Nonetheless, the results are promising, and EM-specific data would be interesting. For what it’s worth, having used an AI-based CXR tool in a resource limited setting, it anecdotally seemed to increase my efficiency and accuracy in identifying abnormal findings. Similar software showing up in your local ED is undoubtedly on the horizon.
Source
Nonradiology Health Care Professionals Significantly Benefit From AI Assistance in Emergency-Related Chest Radiography Interpretation. Chest. 2024 Jul;166(1):157-170. doi: 10.1016/j.chest.2024.01.039. Epub 2024 Jan 29. PMID: 38295950; PMCID: PMC11251081.
