Automating Elastography: CI MED Student Leverages Machine Learning to Streamline Analysis of Brain Tissue

9/22/2024

Written by

This is an example of the type of image being produced in Professor Sutton's lab. A FLAIR image, left, that is conventionally used to show what part of the brain is impacted by the disease, is shown with the corresponding MRE stiffness map from the patient showing stiffness changes due to disease. The affected side is indicated by the arrow.
This is an example of the type of image being produced in Professor Sutton's lab. On the left is a FLAIR image that is conventionally used to show what part of the brain is impacted by the disease. On the right is the corresponding MRE stiffness map from the patient showing stiffness changes due to disease. The affected side is indicated by the arrow.

A Carle Illinois College of Medicine student is using machine learning to automate the painstaking process of generating elastographs – visual maps of stiffness in brain tissue. The new method could enable researchers and doctors to quickly spot increased or decreased brain tissue stiffness associated with a variety of neurological disorders.

When clinicians need to visualize structural changes in a patient’s brain that may indicate disease, they order magnetic resonance imaging (MRI). But increasingly, neurological experts are looking to magnetic resonance elastography (MRE) to provide important clues about brain tissue stiffness and its role in everything from epilepsy to brain tumors and memory deficits. 

<em>John Squire</em>
John Squire

“Our tool will hopefully be used to enhance the pipeline of generating elastographs, which currently takes a long time due to processing of unneeded information (about two to three days for a single elastograph),” CI MED student John Squire said. He recently presented his research into using machine learning to help process elastographs at the IEEE Engineering in Medicine and Biology conference in Orlando.

MRE is a special scan that records MRI data while small vibrations are applied to an organ, creating a visual ‘map’ of tissue stiffness. When imaging the brain, MRE offers researchers and clinicians a new way to analyze variations in flexibility in the body’s most complex organ. “With MRE, we have the potential to ‘palpate’ the brain through imaging, potentially leading us to identify new pathologies or identify them earlier,” Squire said.

The problem is that the MRE process creates artifacts that must be ‘masked’ or removed to create an accurate analysis of the tissue’s characteristics. The current manual MRE masking process requires specialized training and delays the results of the scan significantly. Squire found he could use machine learning techniques to automate the MRE masking process for faster yet reliable results.

<em>Brad Sutton.</em>
Brad Sutton

“With the automated masking that John is doing, we will be able to script the entire workflow, so a person will have an MRE scan, and the stiffness images will be viewable right after the scan,” said CI MED Health Innovation Professor Brad Sutton, who heads the Magnetic Resonance Functional Imaging Lab where the work is based. “The impact is to make the process of getting an MRE exam reproducible, available, and automated to get results out quickly to the clinical team.”

MRE has been used most frequently to noninvasively stage liver cancer, but Sutton and his team are studying the role of stiffness in several different brain disorders. “The most promising applications right now are to see inflammatory responses in the brain, properties of tumors, and to look for small areas of stiffness in the surface of the brain that can be a point where seizures start in epilepsy,” Sutton said. Sutton’s lab previously found that stiffness changes in the hippocampus ­­– the area of the brain involved in memory – may provide early warning signs of developing epilepsy.

Sutton said MREs also hold great promise in helping clinicians develop treatment plans for patients with brain tumors. “Just looking at the tissue structure with MRI does not give any indication as to whether a region is stiff or not, but knowing the stiffness of a tumor, for example, can tell you a lot about its composition, grade, response to treatment, and help plan for its removal,” Sutton explained.

Squire, who is interested in both diagnostic and interventional radiology, plans to continue developing his tool for broader application. “One of our future goals for our project is to eventually develop a stand-alone machine learning application that other researchers might be able to use to isolate the brain in MRE. With this tool, we hope to help others leverage  MRE and move it forward to potentially being used clinically in identifying brain pathology,” Squire said.


Share this story

This story was published September 22, 2024.