Predicting Suicide Risk by Using Neural Representations of Suicide-Related Concepts

Predictions of suicide risk, even using machine-learning methods on clinical data, are quite poor because they rely on patients' self-reports. These researchers used machine-learning analysis to identify stable brain-activation patterns among 17 young suicide ideators and 18 healthy controls who were presented with positive, negative, and suicide-related words (10 each), which were shown multiple times.

Participants were asked to think actively about the concepts. The ideators group was further divided into attempters and nonattempters. Neural representations of six words (death, cruelty, trouble, carefree, good, and praise) in five brain regions associated with self-referential thought distinguished ideators from controls with 0.94 accuracy…