A combination of MR imaging and machine learning may be able to detect individuals who are susceptible to depression before its onset, according to a new study conducted by the University of Texas at Austin.
The researchers are using a supercomputer to train a machine learning algorithm to identify commonalities among hundreds of patients. Brain MR images, genomics data and other factors will help it to accurately predict those at risk of depression and anxiety.
Story Continues Below Advertisement
Accelerate your ROI with our Black Diamond Certified refurbished systems. One year warranty - ISO 13485 Certified - FDA registered - Over 65k parts in inventory
The type of machine learning that the team tested is called Support Vector Machine Learning. They provided a set of training examples that belonged to healthy individuals or those with depression.
The researchers labeled features in the data they considered to be meaningful, and then the examples were used to train the algorithm. A computer scanned the data to find subtle connections between disparate parts, and then constructed a model that assigns new examples to the healthy or depression category.
For the study, the team analyzed brain data from 52 individuals who were seeking treatment for depression and 45 healthy controls. They matched a subset of depressed participants with healthy individuals based on age and gender to compare the groups.
The participants underwent diffusion tensor imaging MR exams. From that data, the team compared fractional anisotropy measurements between the two groups and found statistically significant differences.
The study proved that DTI-derived fractional anisotropy maps can accurately identify those with or at risk of depression versus healthy controls.
The scale and complexity of the research made it impossible for the team to look at each scan, so machine learning was required to automate the discovery process.
"This is the wave of the future," David Schnyer, lead researcher, cognitive neuroscientist and professor at the university said in a statement. "We're seeing increasing numbers of articles and presentations at conference on the application of machine learning to solve difficult problems in neuroscience."
He cautions that even though the results are promising, they're not enough to be used as a clinical metric. But he believes that the algorithm can perform much better by adding more genomic data.
Schnyer and his team will expand their study to include data from several hundred volunteers from the Austin community who have been diagnosed with depression, anxiety or a related condition.
In addition, a new supercomputer will come online at the university later in the year that will be twice as powerful as the current system. That will provide better computer processing power, which is needed to incorporate more data and achieve greater accuracy.