Machine-learning system has high accuracy for suicide risk detection

Paul Basilio, MDLinx | December 07, 2017

In an attempt to predict suicide risk, a research team led by Marcel Just, PhD, at Carnegie Mellon University and David Brent, MD, at the University of Pittsburgh, both in Pittsburgh, PA, employed functional magnetic resonance imaging (fMRI) to look for brain activity signatures in young adults. The findings were published in Nature Human Behaviour.

Advertisement

In previous research, neural signatures associated with the emotions of sadness, shame, anger, and pride differed between people with suicidal thoughts and healthy controls.

The team focused on the networks of activity that represent concepts and evoked emotions. Participants in the study included 38 young adults with current thoughts of suicide and 41 healthy controls.

Participants were placed in an fMRI machine and were shown 30 words for 3 seconds each. The words were related to suicide (eg, “death,” “funeral,” and “hopeless”), positive ideas (“bliss,” “carefree,” and “comfort”), and negative ideas (“boredom,” “gloom,” and “worried”).

Data from 33 participants were used to “train” a machine-learning system to identify differences in networks of brain activity. The system was tested on brain images from 34 participants.

Results showed the system had an accuracy of 91%. It correctly identified 15 of 17 suicidal people and 16 of 17 controls. The strongest differences between the groups were for the words “death”, “carefree,” “good,” “cruelty,” “praise,” and “trouble,” respectively.

In the group of 17 individuals with suicidal thoughts, the system distinguished the 9 who had previously attempted suicide from the 8 who hadn’t with an accuracy of 94%. Discriminating regions were spread across several areas of the brain.

When the system was used on brain images from 21 people with suicidal thoughts with lower quality scan data, it still distinguished the suicidal participants from healthy controls with 87% accuracy.

Prior research has identified brain activity signatures with different emotions. Neural signatures associated with the emotions of sadness, shame, anger, and pride differed between people with suicidal thoughts and healthy controls. The machine-learning system could distinguish between groups based on emotional signatures with an accuracy of 85%.

“People with suicidal thoughts experience different emotions when they think about some of the test concepts,” Dr. Just said. “For example, the concept of ‘death’ evoked more shame and more sadness in the group that thought about suicide. This extra bit of understanding may suggest an avenue to treatment that attempts to change the emotional response to certain concepts.”

Additional research is being conducted to investigate other methods that correlate with the fMRI signatures that may be easier to implement in clinical settings.

Their study was funded by the National Institutes of Health’s National Institute of Mental Health.

To read more about this study, click here

Advertisement