Diagnosing Suicidal Ideation from Resting State EEG Data Using a Machine Learning Algorithm.

Area of Science:

• Neuroscience and Artificial Intelligence
• Clinical Psychology and Psychiatry

Background:

• Suicide is a global health crisis necessitating improved objective risk assessment methods.
• Current interview-based assessments for suicide ideation are often inaccurate.
• Major depressive disorder (MDD) is a significant risk factor for suicidal behavior.

Purpose of the Study:

• To develop a machine learning algorithm (MLA) for predicting suicide ideation.
• To utilize resting-state electroencephalography (EEG) data for objective risk assessment.
• To enhance the accuracy of suicide risk evaluation in individuals with MDD.

Main Methods:

• Collected resting-state EEG data from 224 subjects with MDD.
• Applied a four-step prediction algorithm: EEG preprocessing, brain source localization (ReLORETA), symbolic transfer entropy (STE) for connectivity, and MLA.
• Compared Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbor (KNN) classifiers on balanced datasets.

Main Results:

• All three MLA classifiers achieved high accuracy in predicting suicide ideation.
• Support Vector Machine (SVM) demonstrated the highest performance with 88.9% accuracy.
• Propensity score analysis was used to balance the dataset for reliable comparison.

Conclusions:

• Machine learning analysis of EEG data offers a potential non-verbal method to improve suicide risk assessment.
• Objective, quantitative methods derived from EEG can supplement traditional risk evaluation tools.
• Further research into EEG-based biosignatures could aid in suicide prevention strategies.