Intraventricular Vector Flow Mapping with Data Fusion and Uncertainty Quantification

Left ventricular (LV) flow patterns contribute to diastolic suction, energetic efficiency, and cardiovascular homeostasis. Echocardiographic vector flow mapping (VFM) allows clinicians to non-invasively quantify LV flow nearly in real time. However, despite recent advances, VFM is limited by assuming planar flow and exclusively relying on uncertainty-prone boundary conditions to uniquely define (i.e., regularize) the velocity field. Here, we present a VFM method rooted in Bayesian inference that allows us to flexibly incorporate multi-modality acquisitions (e.g., echo-PIV and Doppler acquisitions and/or multi-velocity encoding Doppler acquisitions) to achieve robust regularization. The algorithm enforces priors based on flow physics and incorporates the uncertainty of the input data. Of note, we include the LV wall boundary condition uncertainty obtained from a deep learning segmentation method. We apply the new VFM method to two ground-truth synthetic velocity fields, one of which is divergence-free while the other is not. We simulated Doppler acquisitions and contrast agent bubbles on these data to perform echo-PIV and feed the VFM algorithm. We also apply the new VFM algorithm to clinical echo-PIV and Doppler acquisitions, including multi-VENC sequences.