Aortic valve leaflet motion for diagnosis and classification of aortic stenosis using single view echocardiography.

Abstract

Background: Accurate classification of aortic stenosis (AS) severity remains challenging despite detailed echocardiographic assessment. Adjudication of severity is informed by subjective interpretation of aortic leaflet motion from the first image parasternal long axis (PLAX) view, but quantitative metrics of leaflet motion currently do not exist. The objectives of the study were to echocardiographically quantify aortic leaflet motion using the PLAX view and correlate motion data with Doppler-derived hemodynamic indices of disease severity, and predict significant AS using these isolated motion data.

Methods: PLAX loops from 200 patients with and without significant AS were analyzed. Linear and angular motion of the anterior (right coronary) leaflet were quantified and compared between severity grades. Three simple supervised machine learning classifiers were then trained to distinguish significant (moderate or worse) from nonsignificant AS and individual severity grades.

Results: Linear and angular displacement demonstrated strong correlation with aortic valve area (r = 0.81 and r = 0.74, respectively). Severe AS cases demonstrated global leaflet motion of 2.1 mm, compared with 3.6 mm for moderate cases (P < 0.01) and 9.2 mm for control cases (P < 0.01). Severe cases demonstrated mean global angular rotation of 11°, significantly less than moderate (18°, P < 0.01) and normal cases (47°, P < 0.01). Using these novel metrics, a simple supervised machine learning model predicted significant AS with an accuracy of 90% and area under the receiver operator characteristics curve (AUC) of 0.96. Prediction of individual severity class was achieved with an accuracy of 72.5% and AUC of 0.88.

Conclusions: Advancing severity of AS is associated with significantly reduced linear and angular leaflet displacement. Leaflet motion data can accurately classify AS using a single parasternal long axis view, without the need for hemodynamic or Doppler assessment. Our model, grounded in biological plausibility, simple linear algebra, and supervised machine learning, provides a highly explainable approach to disease identification and may hold significant clinical utility for the diagnosis and classification of AS.