Deep Learning 4D flow MR Pipeline for the Automated Assessment of Aortic Hemodynamics in BAV

Project Summary
Bicuspid aortic valve (BAV) is the most common congenital heart defect, and it predisposes patients to
complications such as aortic stenosis (AS, most common cause) and aortic aneurysm. BAV patients can
develop highly divergent outcomes (e.g., rapid progressive aortic dilatation vs. no long-term complications) but
the underlying mechanisms that determine the individual risk for complications are not well understood. There
is growing evidence that BAV-based changes in aortic hemodynamics are drivers of aortic wall remodeling and
subsequent aortic dilation. A number of studies by our group and others have shown that 4D flow MRI can
measure altered aortic 3D hemodynamics in-vivo and has potential to provide better assessments of risk for
aortic dilatation in BAV patients. However, current implementations of 4D flow MRI are hampered by long
acquisition times (8-15 minutes) and cumbersome manual processing, such as eddy current corrections, noise
masking, and 3D segmentations. The goal of this project is to develop an deep learning-based acquisition,
image reconstruction, and analyses pipeline for efficient and highly accelerated aortic 4D flow MRI.
The first aim of this proposal is development and validation of a highly accelerated 2-point velocity encoding
4D flow MRI with deep learning reconstruction. This will allow enable a 4D flow sequences with low scan times
(<2 mins) without sacrificing image quality or hemodynamic accuracy. The second aim will be the development
of a deep learning-based automated processing pipeline that will enable rapid processing of aortic
hemodynamics and calculation of the wall shear stress dynamics and relative area changes. In the third aim,
20 BAV patients and 20 healthy controls will be recruited from the patient population at Northwestern, then
imaged and analyzed using the new protocol. This will demonstrate the utility of using the highly optimized
method for the acquisition and analysis of 4D flow MR in a clinical setting. Clinical collaborators will help guide
the project to fulfil the ultimate goal of improving clinical imaging and analysis of these complex patients. And
Siemen support will help with pulse sequence development and the direct integration of our deep learning
network on the scanner, so that this project can be easily integrated in clinical workflows.

Grant Number: 1F31HL168876-01A1
NIH Institute/Center: NIH
Principal Investigator: Haben Berhane