Abstract
Two-dimensional echocardiography (Echo) is a feasible method for assessing left ventricular (LV) ejection fraction (EF) in clinical routine. However, LVEF assessment is dependant on the user’s expertise and is time consuming.
We aimed to compare the accuracy of an artificial intelligence (AI) system in evaluating Echo exams and calculating biplane LVEF against human interpretation and cardiac computed tomography (CT), which was regarded as the gold standard.
A diverse cohort of patients (n=190) with symptomatic heart valve disease underwent both Echo and cardiac spiral CT on the same day. Biplane LVEF was manually traced online on apical 4- and 2-chamber views by experienced cardiologists (Human). After completion of the exam,a novel vendor-neutral AI system performed both automated evaluation of Echo exams, appropriate view selection, and calculations of biplane LVEF in one workflow (Figure 1A). LVEF results from CT were supervised by an experienced radiologist (>10 years) (Figure 1B). Pearson’s correlation coefficient (R), regression analysis, and mean absolute error (MAE) were used to assess the agreement of LVEF values between methods.
The final cohort consisted of 182 patients with complete LVEF assessment by the AI (feasibility: 96%). Although LVEF from AI and Human showed good agreement (R=0.79 and MAE= 5.9), AI exhibited higher agreement with CT (R=0.88, MAE=5.1) than Human LVEF calculations with CT (R=0.79, MAE=5.9), as shown in Figures 1C and 1D.
Author notes
Funding Acknowledgements: None.
