Use of artificial intelligence by novice users to enable accurate point-of-care echocardiographic assessment of left ventricular ejection fraction

Abstract

Abstract Introduction Point-of-care ultrasound (POCUS) is now universal in the hands of non-experts. However, the results are usually binary (normal/abnormal) but not necessarily specific. Artificial intelligence (AI) is currently used by non-experts in different imaging modalities including echocardiography to aid in diagnosis and decision making. Aim To prospectively evaluate whether medical students trained in POCUS and using an AI based assessment tool can accurately evaluate the left ventricular ejection fraction (LVEF) of patients hospitalized in the cardiology department. Methods Nine medical students were trained in a 6-hrs session that included lectures and hands-on practice. Participants used a hand-held ultrasound machine (VScan Extend) equipped with LVivoEF, an AI-based tool that automatically evaluates LVEF. The clips were assessed for LVEF by three methods: visually by the students, students plus the AI-based tool, and experienced cardiologists. All LVEF measurements were compared with the gold-standard (a formal echocardiogram completed within 24-hrs with the Simpson method and LVEF eyeball assessment by two blinded fellowship-trained echocardiographers). Results The study included 88 patients, (aged 58.3±16.3 yrs, mean BMI 28.3±4.4 kg/m2). Comparing LVEF (continuous values) reported by medical students' visual evaluation, medical students plus AI, and cardiologists vs. the echocardiographers values, revealed Pearson correlations of 0.32 (p=0.003), 0.67 (p<0.0001), and 0.73 (p<0.0001), respectively. The agreements between these three evaluation methods and the echocardiographers, assessed at different LVEF categorical thresholds (at 50% and 40%), using kappa coefficient are presented (Figure 1). Conclusion Medical student use of an AI-based tool with a hand-held ultrasound device can improve their LVEF visual assessment to a level of experienced cardiologists. In addition, the use of AI enabled achieving a moderate to substantial inter-rater reliability with echocardiographers' evaluation. This AI tool can be successfully utilized as a decision support tool for POCUS LVEF evaluation by non-experts. Funding Acknowledgement Type of funding sources: None.