Point-of-care AI-enhanced novice echocardiography for screening heart failure (PANES-HF)
Weiting Huang, Tracy Koh, Jasper Tromp, Chanchal Chandramouli, See Hooi Ewe, Choon Ta Ng, Audry Shan Yin Lee, Louis Teo, Yoran Hummel, Feiqiong Huang, Carolyn S.P. Lam
Abstract
The increasing prevalence of heart failure (HF) in ageing populations drives demand for echocardiography (echo). There is a worldwide shortage of trained sonographers and long waiting times for expert echo. We hypothesised that artificial intelligence (AI)-enhanced point-of-care echo can enable HF screening by novices. The primary endpoint was the accuracy of AI-enhanced novice pathway in detecting reduced LV ejection fraction (LVEF) < 50%. Symptomatic patients with suspected HF (N = 100, mean age 61 ± 15 years, 56% men) were prospectively recruited. Novices with no prior echo experience underwent 2-weeks' training to acquire echo images with AI guidance using the EchoNous Kosmos handheld echo, with AI-automated reporting by Us2.ai (AI-enhanced novice pathway). All patients also had standard echo by trained sonographers interpreted by cardiologists (reference standard). LVEF < 50% by reference standard was present in 27 patients. AI-enhanced novice pathway yielded interpretable results in 96 patients and took a mean of 12 min 51 s per study. The area under the curve (AUC) of the AI novice pathway was 0.880 (95% CI 0.802, 0.958). The sensitivity, specificity, positive predictive and negative predictive values of the AI-enhanced novice pathway in detecting LVEF < 50% were 84.6%, 91.4%, 78.5% and 94.1% respectively. The median absolute deviation of the AI-novice pathway LVEF from the reference standard LVEF was 6.03%. AI-enhanced novice pathway holds potential to task shift echo beyond tertiary centres and improve the HF diagnostic workflow.