Machine learning: Trained with MRI

Skip to Navigation


  • Published: Apr 1, 2017
  • Author: David Bradley
  • Channels: MRI Spectroscopy
thumbnail image: Machine learning: Trained with MRI

Scanning for depression

Region of anterior thalamic radiation that revealed significantly lower axial diffusivity and mean diffusivity values in those with major depressive disorder relative to healthy control participants. From the top left clockwise -- coronal, sagittal, 3-D render and axial projections. Credit: David M Schnyer, Peter C. Clasen, Christopher Gonzalez and Christopher G. Beevers

Depression affects millions of people with many undiagnosed. It is now recognised as a leading cause of life-changing debilitation for people between the ages of 15 and 44. US scientists are now using a supercomputer to analysis magnetic resonance imaging (MRI) brain scans to find markers for predicting the illness prior to onset.

David Schnyer, a cognitive neuroscientist and professor of psychology at The University of Texas at Austin, and his colleagues are identifying the characteristics of the brain in the stages before depression arises using the Stampede supercomputer at the Texas Advanced Computing Center (TACC). They have "trained" the system running a machine learning algorithm that can identify common features in hundreds of patients, genomics data and other factors, which they hope will allow them to predict with accuracy the likelihood of depression and anxiety when a new patient presents for an MRI.

Sidestepping the descriptive

Of course, researchers have studied mental disorders by examining the relationship between brain function and structure in neuroimaging data for many years. "One difficulty with that work is that it's primarily descriptive," Schnyer explains. "The brain networks may appear to differ between two groups, but it doesn't tell us about what patterns actually predict which group you will fall into." He and his colleagues are now looking for diagnostic measures that will be predictive of vulnerability to depression and even dementia.

Schnyer has worked on this endeavour with Peter Clasen of the University of Washington School of Medicine, Christopher Gonzalez of the University of California, San Diego, and Christopher Beevers of the University of Texas Austin . The team published their proof-of-concept study in the journal Psychiatry Research: Neuroimaging and report 75 percent accuracy in their predictions with this system.

Machine learning is involves making an algorithm that can "learn" as it is fed sample data and told whether or not its output based on the input is correct or not. Then when it is fed new inputs one can be sure that the outputs will have been guided by the training and allow independent predictions on new data to be made with some confidence.

More data

Schnyer and his colleagues used Support Vector Machine Learning and used MRI scan data from healthy people and those diagnosed with depression. The computer scanned the data, found subtle connections between disparate parts, and built a model that assigns new examples to one category or the other. In their proof of principle, they analyzed brain data from 52 treatment-seeking participants with depression, and 45 healthy control participants. To compare the two, a subset of depressed participants was matched with healthy individuals based on age and gender, bringing the sample size to 50.

The participants were given a diffusion tensor imaging (DTI) MRI scan. Thus, by measuring water diffusion over time, the team could obtain some measure of the integrity of white matter pathways within the cerebral cortex. The team then compared the fractional anisotropy measurements between the two groups and found statistically significant differences. When they reduced the number of voxels involved to a subset that was most relevant for classification and carried out the classification and prediction using the machine learning approach they were able to classify depressed or vulnerable individuals against the healthy controls. The experiments showed how predictive information is distributed across brain networks rather than being highly localized.

"Not only are we learning that we can classify depressed versus non-depressed people using DTI data, we are also learning something about how depression is represented within the brain," explains. "Rather than trying to find the area that is disrupted in depression, we are learning that alterations across a number of networks contribute to the classification of depression."

The team now needs to add more data to better train their algorithm to improve accuracy and to work towards the system becoming viable for real-life clinical diagnosis of depression, anxiety and related disorders.

Related Links

Psych Res 2017, online: "Evaluating the diagnostic utility of applying a machine learning algorithm to diffusion tensor MRI measures in individuals with major depressive disorder"

Article by David Bradley

The views represented in this article are solely those of the author and do not necessarily represent those of John Wiley and Sons, Ltd.

Follow us on Twitter!

Social Links

Share This Links

Bookmark and Share


Suppliers Selection
Societies Selection

Banner Ad

Click here to see
all job opportunities

Copyright Information

Interested in separation science? Visit our sister site

Copyright © 2018 John Wiley & Sons, Inc. All Rights Reserved