Comparison of Artificial Intelligence-Driven Echocardiographic Assessment of Diastolic Function With the ASE...

Contemporary guideline-based assessment of left ventricular diastolic function and filling pressure requires multiple echocardiographic parameters that may not be routinely acquired in real-world clinical practice.

The aim of the study was to compare the prognostic and diagnostic performance of a previously validated deep learning (DL) model for left ventricular diastolic function assessment with those of the 2016 ASE (American Society of Echocardiography)/EACVI (European Association of Cardiovascular Imaging) and 2025 ASE guidelines.

The DL model and the guidelines were compared in the ARIC (Atherosclerosis Risk In Communities) cohort (n = 5,450) for prognostication and in 2 hemodynamic validation cohorts from the United States (n = 83) and Japan (n = 130) for diagnosing elevated left ventricular filling pressure. In all 3 cohorts, ≥1 guideline-recommended parameters were unavailable in a substantial proportion of patients.

In the ARIC cohort, the DL model achieved a higher C-index than both the 2016 ASE/EACVI and 2025 ASE guidelines (0.676 vs 0.638 and 0.602, respectively; both P < 0.001). In the United States cohort, the DL model achieved a higher area under the receiver operating characteristic curve (AUC) than the 2025 ASE guidelines (0.879 vs 0.822; P = 0.041) and a similar AUC to the 2016 ASE/EACVI guidelines (0.879 vs 0.812; P = 0.138). In the Japanese cohort, the DL model achieved a higher AUC than both guideline versions (0.816 vs 0.694 and 0.634, respectively; both P < 0.05).

In settings where not all guideline-recommended parameters are available, the DL model may improve prognostic and diagnostic performance compared with both guideline versions. Nevertheless, these exploratory findings require further validation to confirm that the observed advantages persist when all guideline-recommended parameters are systematically acquired.