Bridging Brain Signals and Self-Reported Symptoms: An AI-Driven, High-Sensitivity Model for Detecting Suicidality in Major Depressive Disorder

BackgroundMajor depressive disorder (MDD) with suicidality represents a significant public health concern, as suicide ranks among the leading causes of death worldwide. While electroencephalography (EEG) has shown promise in depression diagnosis, its utility in identifying suicidal risk remains underexplored. This study aims to develop and validate a Suicidal Risk Index (SR Index) using EEG biomarkers and machine learning (ML) algorithms for distinguishing between MDD patients with and without suicidality. MethodsIn this retrospective observational study, resting-state EEG data were collected using Stress EEG Assessment (SEA) system. SR Index was developed by integrating the PHQ-9 scale with EEG-derived features, including band power, coherence, and fractal dimension, optimized using ML algorithms to enhance accuracy. ResultsThe study included 268 participants (159 without suicidality, 109 with suicidality). The SR Index demonstrated robust discriminative ability with an AUC of 0.8117 (p=2.63x10-18). At the optimal cutoff value of 8, the model achieved 88.99% sensitivity, 57.23% specificity, 67.86% positive predictive value, and 78.85% negative predictive value, with a balanced accuracy of 73.11%. ConclusionsThe SR Index shows promise as an objective tool for identifying suicidality in MDD patients, potentially complementing traditional clinical assessments. This approach may enhance early detection and risk stratification in clinical settings, potentially improving suicide prevention strategies. HighlightsO_LINovel EEG- and ML-based Suicidal Risk Index distinguishes MDD patients with and without suicidality. C_LIO_LISR Index achieves 88.99% sensitivity and 73.11% balanced accuracy in identifying suicidal risk. C_LI