Suicide rates have risen in the U.S. for nearly two decades. But it remains difficult to predict who will attempt suicide, and those at risk may not communicate to others when they experience suicidal thoughts.

Hence the importance of research co-authored by recipient of the 2006 Ruane Prize for Outstanding Achievement in Child and Adolescent Psychiatric Research and 2001 Distinguished Investigator David A. Brent, M.D., and 2012 Young Investigator Lisa A. Pan, M.D., both at the University of Pittsburgh School of Medicine. They reached out to Marcel Just, Ph.D., of Carnegie Mellon University, who pioneered the use of machine learning to identify neural signatures of concepts and emotions using functional magnetic resonance imaging (fMRI) scanning.

These three along with other colleagues conducted research reported last fall in Nature Human Behavior on neural activation patterns of individuals when thinking about words related to suicide, as well as positive and negative concepts.  Working with these patterns, they used machine learning to identify neural signatures of brain activity that were associated with suicidal ideation and behavior.

Seventeen suicidal young adults and 17 healthy controls were studied, and on the basis of the machine learning algorithm derived from brain activation patterns, the suicidal and non-suicidal youth could be distinguished with 91% accuracy. Specifically, the team used fMRI to gauge their participants’ responses to words like “death,” “cruelty,” “trouble,” and “carefree,” finding differential brain activation in brain regions linked to self-referential thinking.

They also found differences between those who had thought about suicide but had never engaged in actual suicidal behavior vs. those who had already attempted suicide, largely based on their responses to the words “death,” “lifeless,” and “carefree,” with a classification accuracy of 94%. Suicidal young adults who had attempted suicide showed evidence of less sadness when thinking about death, for example, compared to those who had considered, but had never attempted suicide.

“This study supports the view that individuals who seriously consider, and who attempt suicide, think about suicide differently than do non-suicidal individuals, and that these differences are reflected in alterations in brain activation patterns while thinking about suicide-related concepts,” Dr. Brent says.

The researchers noted that the program’s ability to identify whether people had actually attempted suicide suggests that it is uniquely sensitive to suicidal risk, and not just to symptoms of related conditions like depression. “This promising pilot study has led the team to develop a larger study, to be funded by NIMH, examining the ability of this approach to predict future suicidal behavior, and to develop a behavioral computer-based task that will capture the neural alternations found during fMRI study, but will be more practical to administer in clinical settings,” Dr. Brent says. In the future, the investigators hope that these tools can be used not only to monitor suicidal risk, but also to frame targets for intervention and prevention.