Suicide is the 10th leading cause of death in the U.S. After a period of decline, it rose 24% in the 15 years ending in 2014, and a gender gap has persisted—four times as many men as women kill themselves.
For clinicians, identifying who is at risk for suicide has long posed a challenge. How do you tell the difference between a person who is distressed but not in danger from someone quietly planning to take his own life? If we could answer that question, we could prevent many untimely deaths. Research shows that taking one’s life is rarely a spur-of-the-moment idea and that most suicidal people have a plan in mind before they act on it.
Now, it seems, computers may be able to help discern who is in danger. A study published last month in the journal Nature Human Behaviour shows that machines can learn to identify suicidal people based on their brain scans.
“Human brains have a common way to represent objects and emotions,” said Marcel Just, a professor of cognitive neuroscience at Carnegie Mellon University and the paper’s first author. That process is so universal, Dr. Just said, that his team’s scanning studies have found the same brain-activation patterns for a word like apple in English, Portuguese and Mandarin speakers.
Four years ago, Dr. Just’s team began capturing brain-activation patterns for emotions by putting actors into brain scanners. Researchers asked them to imagine scenarios that would make them feel anger, envy, shame and other emotions, thus capturing neural signatures for these mental states. The researchers then had a basic visual dictionary of how the brain represents emotions, a resource that would come in handy for their study on suicide risk.
In that study, Dr. Just’s team exposed 34 adults under age 30 to over two dozen words repeated randomly while they were lying in the scanner. Of the subjects, half had a history of suicidal thoughts or suicide attempts. The other half, with no history of mental illness, were the control group.
While in the scanner, the subjects saw slides of emotionally evocative words, including “carefree,” “cruelty,” “praise,” “gloom” and “lifeless.” The participants’ neural responses to these words were carefully mapped using voxel analysis—which captures varying patterns of brain activation according to a 3-D grid of about 20,000 voxels. These are created when neuroscientists electronically dice up the brain into 3-D cubes so they can measure and compare what’s happening in various locations.
Only 120 voxels or so reflect how the brain processes emotional concepts, said Dr. Just, adding: “You show me the activation pattern in those 120 voxels and I’ll show you what word you’re thinking about.”
By analyzing small pattern differences in the neural signatures, the team created a computer algorithm—a set of rules to follow in digital calculations—that could learn to differentiate people with suicidal intentions from members of the control group.
Based on the computer’s prototype of each group’s neural responses, the algorithm could predict whether a subject had previously thought about suicide—or had no such history—with 90% accuracy. The machine could also separate those who had contemplated suicide from those who had really tried it, correctly distinguishing between the two 94% of the time.
“I’m a cognitive psychologist. I used to think that the human mind was for arithmetic, reading and planning where to park your car,” Dr. Just said, remarking on the early preoccupations of cognitive science with pure problem solving. “But when I started to do brain imaging, I saw the networks that become activated” when a person thinks of other people, their intentions or their goals.
Predicting the chances of suicide based on biological markers like brain scans is a monumental achievement. Let’s hope a reliable and affordable version will be available to medicine sometime soon.