Use of Artificial Intelligence by Medical Students to Enable Accurate Point-of-Care Echocardiographic Assessment of Ejection Fraction

Introduction

Point-of-care ultrasound (POCUS) is becoming universal in the hands of physicians as well as medical students. However, the results don’t usually achieve the specificity of an experienced cardiologist.

Artificial intelligence (AI) is now used by non-experts (NE) in different imaging modalities including echocardiography to aid in diagnosis and decision making. This is the preliminary result of a prospective study of POCUS operated by medical students using AI compared to experienced echocardiographers to evaluate the ejection fraction (EF) of patients hospitalized on a cardiology ward.

Methods

Eight students participated in a four-hour didactic session that included lectures and hands-on practice. Study participants used a hand-held ultrasound machine (VScan, GE) equipped with LVivoEF, an application by DiA Imaging Analysis, that uses AI to automatically evaluate EF from the apical 4chamber (A4ch) view.

All measurements were compared to the gold standard (eyeballing by experienced echocardiographers and the Simpson method). The primary end-point was to show that AI enables NE to accurately assess EF.

Results

Thirty patients participated after consent (Mean age: 55.2, mean BMI: 29.0). Comparing continuous EF values reported by Students vs. Experts showed a Pearson correlation of 0.5 (p=0.007) whereas Students plus LVivoEF vs. Experts showed a Pearson correlation of 0.82 (p<0.0001). Correlations were statistically significant (p=0.006, Fischer transformation).

Comparing categorical values (45% as a threshold) of Students vs. Experts showed a Kappa Coefficient of 0.67 (95%CI:0.38-0.96) while Students plus LVivoEF vs. Experts had a Kappa of 0.9 (95%CI0.72-1.00).

Conclusion

Medical student use of LVivoEF with a hand-held ultrasound device had a substantial correlation with experts when evaluating EF values from A4ch view. In addition, the use of LVivoEF enabled achieving near-perfect inter-rater reliability with experts when evaluating categorical EF values. LVivoEF can be used by non-experts as a decision support tool for EF evaluation in POCUS settings.