Development and Clinical Validation of a Web-Based AI System for Multi-View Echocardiographic Videos Analysis

Le résumé fourni par la source

Automated interpretation of echocardiography remains challenging due to inter-observer variability and the limited generalizability of existing deep learning models across different imaging views. This study introduces a secure, web-based clinical decision support system designed to overcome these barriers through Unsupervised Domain Adaptation (UDA). The platform enables simultaneous quantification of Left Ventricular Ejection Fraction (LVEF) and wall thickness across standard apical and parasternal views without requiring site-specific retraining. Distinctively, the system features an interactive interface that allows clinicians to visualize and validate segmentation masks in real-time, ensuring transparency in automated diagnostics. Clinical validation involving cardiologists and residents demonstrated high computational efficiency, with an average processing time of 1.15 seconds per cardiac cycle. Statistical analysis revealed a strong agreement with expert manual measurements ($\mathrm{r}=0.95, \mathrm{P}<0.001$) and a negligible mean bias. Furthermore, usability assessments indicated high user satisfaction (6.18/7), with physicians accepting 86% of automated outputs without modification. These findings suggest that integrating robust domain-adaptive algorithms into an intuitive web environment significantly enhances workflow efficiency and fosters clinical trust, paving the way for broader adoption of AI in routine cardiac care.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.