Close to five years ago, Just delivered a talk on the neural signatures of emotion at the University of Pittsburgh. Afterward, a psychiatrist approached him and described his profession's sorry record of predicting suicide. He asked: Could neural signatures help reveal intent?
Just and his co-authors set about devising an approach for the assessment of suicide risk. They would use machine learning to detect abnormal emotional responses to concepts such as "death" and "cruelty," as well as to words such as "carefree" and "good."
In a group of 34 young adult subjects, the resulting program distinguished between healthy controls and suicide-contemplators with an accuracy of 91 percent. It correctly identified 15 of the 17 suicidal participants and 16 of the 17 non-suicidal controls.
A further iteration of the machine-learning program was able to distinguish, with 87 percent accuracy, between subjects who had engaged in suicidal thinking only and those who had attempted suicide.
The activation patterns inside the brains of young adults who had stared into that psychological abyss and acted on the impulse tended to respond to death-related words with less sadness than did subjects who had contemplated suicide but never made an attempt.
Compared to subjects with a past suicide attempt, those who had pondered suicide but not acted on such thoughts responded to death- and suicide-related words like "lifeless," "desperate," "overdose" and "funeral" with neural signatures suggesting more anger, and they did so reliably.
Just acknowledged that, in many cases, the breadth and depth of a subject's depressive symptoms also can predict whether he will try to harm himself. Administering a dynamic brain scan, however, may offer earlier warning that self-destructive thought patterns are settling in, he said.
Understanding how those thought patterns manifest themselves as brain-activation patterns might also offer a way to target psychological therapies, and test whether they are working.
"Obviously it's good to ask the person," Just said. "We don't try to set this up as a competing measure to existing methods, but a complementary one. These are pretty high accuracies we're getting."
(c)2017 Los Angeles Times
Visit the Los Angeles Times at www.latimes.com
Distributed by Tribune Content Agency, LLC.