Predicting Adverse Outcomes for BAV Aortopathy Patients by Fluid Physics-Informed Velocity Estimation from Contrast-Enhanced MR Angiography

Motivation: Aortic hemodynamics have involvement in bicuspid aortic valve (BAV) aortopathy progression, but measuring velocities with 4D flow MRI is time-consuming and not routine clinically. Goal(s): We aimed to evaluate a prognostic value of hemodynamics derived from velocities estimated by a machine-learning network from common standard-of-care CEMRA images. Approach: A cycleGAN was trained to estimate velocity from paired 4D flow/CEMRA training data. Adverse outcomes (surgery/aneurysm/dissection) were tallied for a separate BAV cohort with CEMRA acquired, and the cycleGAN was applied to estimate velocities. Performance predicting outcomes from hemodynamic estimates was evaluated. Results: Good predictive performance (AUC>0.7) using CEMRA-derived hemodynamic estimates was observed. Impact: Strong performance predicting adverse outcomes with estimates of hemodynamics derived from easy-to-acquire CEMRA images suggests high potential clinical utility in management of BAV aortopathy, and this may offer new and powerful ways to stratify risk in patients with BAV aortopathy.