Machine learning uses neuroimaging data to accurately diagnose depression

Written by Jonathan Wilkinson

Researchers from University of Texas at Austin (TX, USA) have used a supercomputer to train a machine-learning algorithm to identify signs of depression from neuroimaging data. The algorithm is able to recognize commonalities among hundreds of patients and classify them as being either healthy or showing features of mental illness. The findings were published in Psychiatry Research: Neuroimaging, and demonstrated that the machine-learning approach could classify individuals with major depressive disorder with roughly 75% accuracy.
Neuroimaging data has long been used to study mental illness, particularly by looking at the relationship between brain function and structure. David Schnyer (University of Texas at Austin), lead author of the study, commented on the limitations of this approach: “One difficulty with that work is that it’s primarily descriptive. 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. We’re looking for diagnostic measures that are predictive for outcomes like vulnerability to depression or dementia.”

The investigators used Support Vector Machine Learning to train the algorithm; examples were provided that had been marked as either healthy individuals or those who had been diagnosed with depression. Additionally, the team labelled features in their data that were meaningful. The computer then scanned the data and built a model that assigns new examples to one of the two categories.

A sample of 50 participants (healthy controls matched with depression patients) received diffusion tensor imaging (DTI) MRI; a comparison of fractional anisotropy measurements was made between the two groups, and a statistically significant difference was found. The researchers then reduced the number of voxels (3D cubes that represent either structure or neural activity throughout the brain) involved to a subset that was most relevant for classification, and then used the machine learning approach to carry out classification and predication.

It was found that DTI-derived fractional anisotropy maps can accurately classify depressed or at-risk individuals compared with healthy controls. Furthermore, information predictive of depression is distributed across brain networks, rather than being localized.

Christopher Beevers (University of Texas at Austin), another of the study authors, explained further: “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. 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.”

Detecting complex relationships among such a vast number of brain components is practically impossible and therefore necessitates a machine learning approach. Schyner explained: “This is the wave of the future. We’re seeing increasing numbers of articles and presentations at conference on the application of machine learning to solve difficult problems in neuroscience.”

In order to move this machine learning approach to the clinic, the researchers believe that more data is needed; not just MRI scans but also from genomics and other classifiers. Schyner and Beevers also plan to expand the study to include data from several hundred volunteers who have been diagnosed with depression, anxiety or a related condition. Later in 2017, they will also have access to a supercomputer that is twice as powerful as the one used in the latest study, allowing the incorporation of more data as well as greater accuracy.

Beevers concluded: “One of the benefits of machine learning, compared to more traditional approaches, is that machine learning should increase the likelihood that what we observe in our study will apply to new and independent datasets. That is, it should generalize to new data. This is a critical question that we are really excited to test in future studies.”

Sources: Schnyer DM, Clasen PC, Gonzalez C, Beevers CG. Evaluating the diagnostic utility of applying a machine learning algorithm to diffusion tensor MRI measures in individuals with major depressive disorder.  Psychiatry Res. Neuroimaging doi:10.1016/j.pscychresns.2017.03.003 (2017) (Epub ahead of print);