CTA-derived hemodynamics using 4D flow MRI and deep learning: A key step towards broad implementation of in vivo aorta hemodynamic quantification for risk-stratification and treatment planning.

4D flow MRI allows for in vivo quantification of thoracic aorta hemodynamics. Our group has been a leader in identifying 4D flow derived parameters that correlate with increased risk of adverse outcomes in many aortic pathologies including bicuspid aortic valve (BAV) aortopathy, aneurysm, coarctation, post-surgery, and aortic dissection. However, 4D flow is associated with several key limitations limiting widespread adoption. Specifically, the technique is not widely available, requires special expertise for acquisition, post-processing, and quantification, has relatively long scan times (10-15 minutes), and surgeons general prefer planning interventions based off of CT angiography (CTA). In this proposal, we will address these limitations by leveraging our dataset of >2500 patients and controls with aortic 4D flow MRI to develop a deep learning-based tool that quantifies hemodynamic parameters directly from CTA. Currently, we have a deep learning-based tool that provides rapid segmentation of the aorta from 4D flow MRI data and a pipeline for creating voxel-wise maps of peak velocity, forward and reverse flow, kinetic energy, and WSS from these segmentations. We will improve this segmentation tool to accurately segment the aorta from CTA. Then, using deep learning generative adversarial networks, we will create an algorithm that can predict the voxel-wise maps directly from segmented CTA data. This tool will be trained, validated, and tested in the subgroup of our 4D flow cohort that also has been imaged with CTA. As a proof-of-concept study, we will correlate CTA-derived peak-velocity and wall-shear stress with the ascending aorta growth rate in a retrospectively identified group of BAV patients. Measuring aorta hemodynamics from CTA will make these parameters accessible to many more sites and patients. It will also significantly broaden the impact and expand our knowledge of the role of in vivo hemodynamic assessment on risk-stratification and treatment planning for numerous aortic diseases.