Predicting how a particular patient's vascular system with respond to different treatment or stimuli and adapt over long periods of time remains a grand challenge in precision medicine. The lack of real-time turn around critically limits our ability to search a wide treatment space to identify optimal intervention plans based on long-term, personalized predictions. Moreover, it prevents real-time monitoring of a patient's hemodynamics based on streaming, dynamic data such as that acquired from wearables. By moving from simulations that can capture only several heartbeats to modeling months or even years, we shift the utilization of patient-specific digital twins to provide on-demand tracking of a patient's hemodynamic state. Such data would improve screening for cardiovascular disease, improved monitoring, and finally, inform treatment planning by enabling prediction of longterm flow effects currently not attainable. The major objective of this proposal is to develop and apply a methodology coupling physics-based simulations with machine learning that, combined with wearable sensors, provides real-time, personalized predictions of 3D, complex hemodynamic patterns over months to years. A better understanding of how a patient's circulatory system and underlying hemodynamics responds under different physiological states over time is of broad relevance to treating a wide range of vascular diseases.

Narrative/Public Health Relevance The importance of understanding hemodynamic profiles in treating vascular disease is now well established, and techniques to derive key parameters at single time points are rapidly being adopted in the clinic. Despite this, the field lacks coherent, realtime, and non-invasive methods to identify 3D profiles over time, or how they can be used to guide treatment. The proposed project will develop a novel, physiologically-aware learning method by which 3D flow fields can be derived on a patient-specific bases, and thus, has broad implications for vascular disease therapy