AI-Assisted Lung Sliding Detection in Point-of-Care Ultrasound by Marine Corps Corpsmen: A Multi-Reader Study

Area of Science:

• Medical Ultrasound
• Artificial Intelligence in Medicine
• Point-of-Care Diagnostics

Background:

• Lung ultrasound (LUS) interpretation faces challenges in austere settings due to training limitations and operator variability.
• Artificial intelligence (AI) offers potential solutions to improve LUS use and diagnostic consistency.
• This study investigated AI's impact on novice users' ability to detect absent lung sliding, a key pneumothorax indicator.

Purpose of the Study:

• To evaluate AI-based decision support for improving diagnostic accuracy and confidence in lung ultrasound (LUS) interpretation.
• To assess the performance of novice military medics in identifying absent lung sliding with and without AI assistance.
• To determine if AI can enhance the reliability of LUS in resource-limited environments.

Main Methods:

• A pilot-prospective, multi-reader, multi-case study involving five novice military medics.
• Each medic interpreted 50 LUS video clips twice: once with AI assistance (ATLAS) and once without, in randomized order.
• Diagnostic performance was measured by area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, and accuracy, with expert consensus as the reference standard.

Main Results:

• AI assistance significantly improved diagnostic performance, increasing mean AUROC from 0.72 to 0.93 (P=.03).
• Sensitivity, specificity, and overall accuracy saw substantial improvements with AI (P<.001).
• Reader confidence ratings significantly improved, with high-confidence ratings nearly doubling and low-confidence ratings decreasing (P<.001).

Conclusions:

• AI-based decision support markedly enhanced diagnostic accuracy and confidence for novice lung ultrasound (LUS) interpretation.
• Real-time AI support shows promise for improving access to high-quality LUS in military and resource-limited settings.
• AI can help overcome training limitations and inter-operator variability in point-of-care ultrasound applications.