Abstract
AIMS: Echocardiography is a rate-limiting step in the timely diagnosis of heart failure (HF). Automated reporting of echocardiograms has the potential to streamline workflow. The aim of this study was to test the diagnostic accuracy of fully automated artificial intelligence (AI) analysis of images acquired using handheld echocardiography and its interchangeability with expert human-analysed cart-based echocardiograms in a real-world cohort with suspected HF. METHODS AND RESULTS: In this multicentre, prospective, observational study, patients with suspected HF had two echocardiograms: one handheld portable and one cart-based scan. Both echocardiograms were analysed using fully automated AI software and by human expert sonographers. The primary endpoint was the diagnostic accuracy of AI-automated analysis of handheld echocardiography to detect left ventricular ejection fraction (LVEF) ≤40%. Other endpoints included the interchangeability (assessed using individual equivalence coefficient [IEC]), between AI-automated and human analysis of cart-based LVEF. A total of 867 patients participated. The AI-automated analysis produced an LVEF in 61% of the handheld scans and 77% of the cart-based scans, compared to 76% and 77% of human analyses of the handheld and cart-based scans, respectively. The AI-automated analysis of handheld echocardiography had a diagnostic accuracy of 0.93 (95% confidence interval [CI] 0.90, 0.95) for identifying LVEF ≤40% (compared to the human analysis of cart-based transthoracic echocardiography scans). AI-automated analysis of LVEF on handheld devices was interchangeable with cart-based LVEF reported by two expert humans (IEC -0.40, 95% CI -0.60, -0.16). CONCLUSIONS: Artificial intelligence-automated analysis of handheld echocardiography had good diagnostic accuracy for detecting LVEF ≤40%. AI-automated analysis of LVEF of handheld scans was interchangeable with cart-based expert human analysis.