Diagnostic accuracy and clinical utility of AI-assisted transthoracic echocardiography for detecting left ventricular systolic dysfunction: a meta-analysis and diagnostic test accuracy

Background Accurate assessment of left ventricular systolic dysfunction (LVSD) is central to cardiovascular diagnosis and management. Artificial intelligence (AI)–assisted echocardiography has been proposed to improve reproducibility and efficiency, but reported diagnostic performance remains heterogeneous. Objective To systematically evaluate the diagnostic accuracy of AI-assisted transthoracic echocardiography (TTE) for detecting LVSD. Methods A systematic review and diagnostic test accuracy meta-analysis was conducted in accordance with PRISMA-DTA guidelines and registered on PROSPERO. MEDLINE, Embase, and trial registers were searched for studies evaluating convolutional neural network–based echocardiographic systems in adults. The index test was AI-assisted echocardiography, and reference standards included expert human-interpreted TTE or cardiac magnetic resonance imaging. Risk of bias was assessed using QUADAS-2. Pooled sensitivity and specificity were estimated using a bivariate random-effects model. Prespecified sensitivity analyses restricted inclusion to cart-based TTE studies and to studies using a left ventricular ejection fraction (LVEF) threshold <50%. Results Seven studies comprising 2,397 participants met inclusion criteria. Two studies evaluated handheld point-of-care ultrasound and five evaluated cart-based TTE. Using a bivariate random-effects model, pooled sensitivity was 0.82 (95% CI 0.73–0.89) and pooled specificity was 0.89 (95% CI 0.82–0.93). Restriction to cart-based TTE yielded pooled sensitivity of 0.80 (95% CI 0.67–0.89) and specificity of 0.91 (95% CI 0.83–0.95). Restriction to studies using an LVEF <50% threshold increased pooled sensitivity to 0.85 (95% CI 0.75–0.91). Risk of bias was generally low, with some concerns related to patient selection. Conclusions AI-assisted echocardiography demonstrates high diagnostic accuracy for detecting LVSD, comparable to expert interpretation. These findings support its role as a diagnostic adjunct, particularly in high-throughput and resource-limited settings.