Illinois researchers develop a new method to efficiently reconstruct MRI images
Magnetic Resonance Imaging (MRI) relies on a strong background magnetic field to image the human body and, ideally, would require this background field to be uniform for accurate localization of signals from different parts of the body. However, achieving uniformity is nearly impossible in practice because different tissues in the body have different magnetic properties and cause variations in the magnetic field. This variation is known as B0 inhomogeneity and is a universal problem in many MRI experiments.
Bioengineering professors Fan Lam and Brad Sutton have been working to develop models that can mitigate the impact of these variations on images and improve the accuracy of localizing and measuring signals in the brain. They published their findings in Magnetic Resonance in Medicine. Lam and Sutton are affiliates of the Beckman Institute for Advanced Science & Technology.
Traditionally, reconstruction models to mitigate the B0 inhomogeneity problem require the incorporation of MRI physics equations that are very complicated and pose a significant computational challenge. "In fact, the problems are so large, that we cannot even store the equations for the problem in memory for any computer that we have access to, we have to compute relationships as we go, based on the physics of the MRI scanner," said Sutton. The process may take a week to reconstruct one set of images on a high power workstation.
"Our paper leveraged a different way to capture a majority of the information contained in those equations with just a few calculations," said Sutton, technical director of the Biomedical Imaging Center at Beckman.
Lam and Sutton identified which encoding element of the model can be approximated and still result in high quality imaging. By identifying the major characteristics of the B0 inhomogeneity variations in the brain, they can establish a few key assumptions that allow a low-rank approximation of the model. This new method dramatically reduces the memory usage and the computation complexity by three orders-of-magnitude.
"It enables us to do things that would not have been possible before, such as going to higher resolutions, more time points in a data time series, or to incorporate additional components of the MRI physics for better-reconstructed images," noted by Sutton and Lam.
"The method can be applied to a range of quantitative MRI acquisitions that are not limited to the brain," said Lam. "Being able to handle B0 inhomogeneity effectively can impact many applications. One of them in particular is MR spectroscopic imaging (MRSI)" Lam’s MRSI research has earned him a National Science Foundation CAREER award this year as his lab continues to study molecular-level activities in the human brain noninvasively.