Presentation

Physics-Informed Functional Priors for Blood Flow Predictions in Data-Poor Regimes
Presenter
DescriptionTranscranial Doppler (TCD) ultrasound is one of the most common techniques in the current clinical workflow that enables noninvasive and instantaneous evaluation of blood flow velocity within the cerebral arteries. This type of prediction has many clinical applications for diagnosing various diseases. However, the number of TCD measurements is limited in clinical settings. Training a deep neural network in these settings is challenging due to the limited data measurements. Therefore, developing a surrogate model that provides predictions based on sparse data is vital. To this end, we present a physics-informed regression framework to predict blood flow properties via very few sparse measurements. Our approach is based on learning the functional priors from one-dimensional computational fluid dynamics (CFD) models. We develop a low-rank approximation methodology to extract functional priors from the CFD, which enables real-time blood flow reconstruction from sparse and noisy measurements. We examine our methodology on examples such as synthetic data of a Y-shaped bifurcation vessel as well as abdominal aorta and brain vasculature.
TimeTuesday, June 2711:30 - 12:00 CEST
LocationSanada I
Event Type
Minisymposium
Domains
Life Sciences