Neural activity during inhibitory control predicts suicidal ideation with machine learning.

Area of Science:

• Neuroscience
• Psychiatry
• Biomarker Discovery

Background:

• Suicide is a major global health concern, with suicidal ideation (SI) as a key risk factor.
• Current suicide prevention strategies often rely on identifying and treating risk factors, but scalable biomarkers for SI are lacking.
• Previous biomarker research for SI has been limited by the reliance on expensive neuroimaging techniques.

Purpose of the Study:

• To investigate the classification of suicidal ideation (SI) using machine learning (ML) on electroencephalography (EEG) data.
• To identify potential EEG-based biomarkers for predicting SI that are clinically scalable and affordable.
• To assess the predictive performance of ML models trained on EEG features during cognitive tasks.

Main Methods:

• Collected EEG data from 76 participants (38 with SI, 38 without SI), matched for age, sex, and mental health symptoms.
• Recorded EEG at rest and during four cognitive tasks: inhibitory control, interference processing, working memory, and emotion bias.
• Analyzed EEG signals in theta, alpha, and beta frequency bands, performing cortical source imaging to identify neural predictors for ML models.

Main Results:

• The best ML model utilized beta band power during an inhibitory control task, achieving 89% sensitivity and 98% specificity for SI classification.
• Shapley analysis identified feedback-related power in visual and posterior default mode networks, and response-related power in ventral attention, fronto-parietal, and sensory-motor networks as key predictors.
• External validation in an independent sample of depressed individuals showed 50% sensitivity and 61% specificity.

Conclusions:

• The study demonstrates a promising, scalable EEG-based approach for predicting suicidal ideation (SI).
• This EEG biomarker may aid in suicide risk identification and intervention strategies.
• Further research is warranted to refine and validate this approach in diverse clinical populations.