Abstract 15710: Point-of-Care Ultrasonography: Artificial Intelligence in the Assessment of Left Ventricular Function

Abstract

Introduction: Point-of-care ultrasonography (PoCUS) has become routine for the bedside assessment of patients. Current evidence shows that image acquisition and interpretation of left ventricular function can be achieved reliably by trained users, but evidence is lacking for interpretation by artificial intelligence. We sought to evaluate the accuracy of PoCUS devices with integrated artificial intelligence in comparison to transthoracic echocardiograms for assessment of left ventricular ejection fraction (LVEF). Methods: This is a prospective, multicenter cohort study using the KOSMOS portable ultrasound device (EchoNous, Redmon, WA). PoCUS studies included an apical four and two chamber views and were obtained by either novice trainees or experienced ultrasonographers. A real-time, automated LVEF was calculated by the device using a modified Simpson’s method. These patients also underwent a formal transthoracic echocardiogram, with image acquisition and interpretation completed by blinded ultrasonographers and level-3 trained echocardiographers, respectively. Results: A total of 449 patients were enrolled in the study with 227 and 222 studies completed by novice and experienced scanners, respectively. Images of suitable quality were obtained in 208 of the novice cohort (91.6%) and 216 of the experienced cohort (97.3%). In comparison to formal echocardiography, linear regression shows a R 2 value of 0.82 for the total cohort, and 0.85 for the novice and 0.72 in experienced scanner subgroups. Incorporating an intra-observer variability of ± 5%, the sensitivity and specificity for an abnormal (≤50%) and severely reduced (≤30%) LVEF was 99.2% and 97.7%, and 86.8% and 99%, respectively. Conclusions: Our study is one first to assess the accuracy of PoCUS devices with integrated artificial intelligence. Based on our study, a real-time, automated LVEF can be acquired by PoCUS with strong correlation to formal echocardiography. This study also highlights that diagnostic quality studies can be acquired by novice scanners with comparable accuracy to experienced scanners. This provides evidence for an immediate and inexpensive method for accurate LVEF assessment in emergent and resource-limited environments.