Artificial Intelligence Guided Nonexpert Echocardiogram in the COVID-19 Health Action Response in Marines 2.0 Study

Purpose: Echocardiography is a widely utilized cardiac imaging modality, but accessibility can be limited by cost and lack of skilled sonographers. We demonstrate the use of point-of-care ultrasound (POCUS) with an embedded deep learning algorithm to guide novice users in obtaining diagnostic-quality echocardiographic images and ejection fraction (EF) estimates. Methods: Utilizing an AI-assisted echocardiography technology among a cohort of young healthy adults, we evaluated 10 examiners on their ability to capture four POCUS cardiac views per participant and calculate a real-time AutoCapture ejection fraction. We assessed the number of studies completed, image quality as defined by quality meter score (QMS), and the acquisition time required per study. Results: Examiners obtained 887 echocardiograms from 789 participants, most of whom were healthy, white (70.3%) males (92.1%) with a median age of 18 years (range 18–34), and an EF of 55%–70% (range 21%–70%). Examiners, categorized as “Beginner,” “Intermediate,” and “Advanced” proficiency, obtained an AutoCapture EF in 69.6%, 70.6%, and 79.1% of studies, and a mean QMS of 71.0, 72.2, and 73.8, respectively, regardless of the view type examined. The mean QMS was highest for parasternal long axis (77.1) compared to the other three views, with no significant difference between the number of studies performed and the QMS for each view (p > 0.050). Conclusions: We demonstrate that novice examiners can utilize this technology to obtain interpretable cardiac images in a timely fashion, and POCUS could be used to identify cardiac conditions in resource-limited settings.