Fusion of computational fluid dynamic flow data into 4D flow MRI using machine learning: September 9, 2020



Speaker: Kevin M. Johnson, PhD, UW-Madison Medical Physics and Radiology

Investigative Team: David Rutkowski, Alejandro Roldán-Alzate, Shiva Rudraraju, Kevin M. Johnson

Date: Wednesday, September 9, 2020

Time: 2:00PM – 3:00PM Central Time

Location: Web conference (a link will be sent out)

Title:Fusion of computational fluid dynamic flow data into 4D flow MRI using machine learning


Synopsis: MRI has the capability of measuring blood flow velocities in the heart and vasculature, with recent 4D Flow techniques providing 3D dimensional visualization of complex flow patterns. 4D Flow MRI – derived metrics have the potential to improve diagnosis and treatment planning of cardiovascular disease; however, scan time restrictions limit MRI flow resolution, and inherent flow encoding properties lead to imperfections in the flow field. As an alternative, velocity fields can be derived from physics constraints using simulation based computational fluid dynamics (CFD); which may address some of these limitations, providing high resolution, low noise velocity fields which satisfy the physics of fluid flow (conservation of mass and momentum). Yet, standalone CFD is also limited due to its dependence on patient-specific input conditions, and truncation errors which propagate from inaccurate assumptions. A method that utilizes the best of both MRI and CFD may allow for enhanced flow analysis but has been challenging to achieve due to the disparity between these techniques. Therefore, the purpose of this work was to develop a machine learning paradigm which fuses information from 4D Flow MRI and CFD using supervised learning, to provide high resolution, physics-based, patient-specific flow fields. We will discuss the augmentation of 4D Flow MRI data with CFD using informed training of neural networks to produce highly accurate physiological flow fields. 

Description of speaker’s research: In the past several decades, major advances have been made in all of the commonly utilized medical imaging modalities such that imaging now plays an integral role in disease diagnosis, treatment planning, and therapy monitoring. With rapid advances in disease etiology, imaging must continue its evolution into a more quantitative, functional, and cellular tool. This is particularly true of MRI which for all its potential, and despite tremendous hardware improvements, remains constricted by it slow speed, low detection sensitivity, and imaging inaccuracies. Due to these factors, MRI is often not utilized for diseases in which it could provide superior diagnostic and prognostic information, and is rarely used for quantitative imaging. It is my aim to enable the potential of MRI by accelerating acquisition speed, removing ambiguities and artifacts, and providing novel techniques for disease quantification. Through these advances, I aim to help develop a new era of quantitative and molecular imaging across a broad spectrum of diseases. Some research interests include:

  • MR pulse sequence development, non-Cartesian and non-Fourier imaging
  • Signal encoding and decoding, sampling theory, recovery from incomplete samples
  • Motion robust imaging, free breathing MRI, motion sensing
  • Macro and micro vascular remodeling, perfusion, flow
  • Improving the MRI experience for patients