MRI Predicts Suicidality with 91% Accuracy / by F. Perry Wilson

Death. Cruelty. Trouble. Carefree. Good. Praise. Using just those 6 words, and a brain’s response to them, researchers were able to identify suicidal individuals with 91% accuracy.

For the video version, click here.

It’s a rather macabre success for a machine-learning algorithm, but the implications are fairly profound.

We’re talking about this study, appearing in Nature Human Behaviour.

Up until now, identifying patients at risk of suicide basically involved, well, asking them if they were thinking about suicide. But prior studies have shown that nearly 80% of patients who committed suicide denied suicidal ideation in their last contact with a mental health practitioner. That we could get at this crucial information using an MRI scanner is potentially game-changing.

But let’s talk about how this study worked.  Researchers identified 17 adults with suicidal ideation and 17 controls.  They put them in an fMRI scanner to measure which parts of the brain would be activated when the subjects were thinking about those key words.

 

Words like death and cruelty differentially activated the left superior medial frontal area and the medial frontal/anterior cingulate in the individuals with suicidal ideation – these are areas associated with self-referential thought. Using a machine-learning algorithm the researchers successfully identified 15 of the 17 brains from those with suicidal ideation and 16 of the 17 controls.

"I'm sorry things are so difficult. Please lie in this claustrophobia-inducing tube for 30 minutes".

"I'm sorry things are so difficult. Please lie in this claustrophobia-inducing tube for 30 minutes".

Of course, the elephant in the room here is a 70,000 pound electromagnet.

 

These patients volunteered to participate in this study, they were able to concentrate for 30 minutes straight as various emotionally-charged words were presented to them. This is not a test we’ll be administering in your therapist’s office anytime soon. I raised that issue with Dr. Marcel Just, the lead author of the study.

 

Marcel Just, D.O Hebb Professor of Psychology, Carnegie Mellon University.

Marcel Just, D.O Hebb Professor of Psychology, Carnegie Mellon University.

“It would be nice to see if we could possibly do this using EEG, if we could assess the thought alterations with EEG. It would be enormously cheaper. More widely used.”

The other thing to remember is that these volunteers told us they had suicidal ideation. That’s the gold-standard that the computer was learning on. But, as I noted above, people who admit they are having suicidal thoughts are the easy ones – we need ways to figure out who is suicidal and not telling us. Dr. Just pointed out that mind-reading, so to speak, is a long way off for a simple reason:

“If somebody didn’t want others to know what they are thinking, they can certainly block that method. They can not coorperate. I don’t think we have a way to get at people’s thoughts against their will.”

What this study does tell us is not just that there are differences in the brains of suicidal individuals and controls, but that those differences are discoverable with technology. Today’s multi-million dollar MRI scanner is tomorrow’s EEG and the next day’s ubiquitous IPhone attachment. But for now, we may need to change an old adage: the fMRI is the window to the soul.