A digital biomarker for aortic stenosis development and progression using deep learning for two-dimensional echocardiography.

In this multi-center cohort study of 12,609 patients with no, mild or moderate aortic stenosis (AS), we explored whether a deep learning-enhanced method that relies on single-view, two- dimensional videos without Doppler can stratify the risk of AS development and progression. Video-based phenotyping based on the digital AS severity index (DASSi) identified patient subgroups with distinct echocardiographic and clinical trajectories independent of the baseline AS stage and profile. The results were consistent across two geographically distinct cohorts and key clinical subgroups, supporting the use of deep learning-enhanced two-dimensional echocardiography as a supplement to the traditional assessment of AS in the community.