Validation of a fully automated workflow for longitudinal strain using artificial intelligence

Manual interpretation of echocardiographic data for strain analysis is time-consuming and prone to inter-observer variability. With the advent of artificial intelligence (AI), its potential to streamline echocardiographic interpretation and reduce variability has attracted increasing interest. This study aimed to evaluate the accuracy of AI-derived strain measurements, focusing on left ventricular global longitudinal strain (LVGLS) and left atrial reservoir strain (LASr), compared with expert sonographer measurements.

A total of 150 patients were included regardless of underlying disease. Echocardiographic parameters, including LVGLS and LASr, were independently assessed by an expert sonographer and the fully automated software (Us2.ai). Correlations and agreement were evaluated using intraclass correlation coefficients (ICC) and complementary analyses. Subgroup analyses were also performed for LVGLS stratified by left ventricular ejection fraction (EF) categories, and for LASr stratified by left atrial volume index (LAVi) categories. For LVGLS, the ICC was 0.83 (95% CI: 0.77-0.88) and 0.77 (95% CI: 0.67-0.84) for LASr. The accuracy of both LVGLS and LASr was generally maintained across different EF and LAVi subcategories.

The fully automated software showed reliable correlations with expert sonographer measurements, particularly for LVGLS, while correlation for LASr was not as high. With further investigation, the fully automated software may become a useful tool for echocardiographic analysis in clinical practice.