Back to search

McMaster University

Suicide Ideation Detection using Advanced Signal Processing and Machine Learning Analysis of the EEG

Abstract

dc:description.abstract

The electrical activity of the human brain provides insight into its complexities, which can be helpful in identifying early signs of neurological or psychiatric problems. These signals reveal underlying functional processes and often show diagnostic potential even before clinical symptoms appear. Electroencephalography (EEG) is commonly used to capture brain signals due to its non–invasiveness, high temporal resolution, and cost–effectiveness. The primary focus of this thesis is to develop novel signal processing and machine learning frameworks that use brain connectivity measures and source separation techniques to detect suicidal ideation from event–related potential (ERP) data. In the first part, we model dynamic brain networks using extended multivariate autoregressive (eMVAR) analysis and extended partial directed coherence (ePDC), taking into account the instantaneous mixing effect that has been neglected with traditional approaches. We estimate time–varying connectivity patterns in the frequency domain from multi–trial ERP data across both standard and deviant stimuli. Then, feature selection is used to reduce dimensionality. Afterwards, employing graph neural networks (GNNs) on the selected feature set resulted in robust performance in distinguishing suicidal individuals from healthy controls. In the second part, we explore the estimation of the instantaneous mixing matrix. Different techniques, including nonlinear optimization solvers, REINFORCE–based optimization, and convexified approaches, were compared. Our results showed that nonlinear solver outperformed alternative methods for both synthesized and real ERP data, confirming the effectiveness of our initial proposed approach. We further introduce a convolutive mixing modelling framework that captures latent neural dynamics and extracts demixing coefficients as features. We propose a dual–branch, permutation–invariant deep learning architecture, which processes both standard and deviant ERPs, and also compare it with other permutation–invariant approaches. Our proposed method achieves high classification performance. In summary, this thesis develops new methods for EEG/ERP analysis that improve the detection of suicidal ideation and thus makes a contribution to addressing the critical, worldwide problem of suicide.

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Torabi, Ali

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
McMaster University
Base URL
macsphere.mcmaster.ca/server/oai/request
Last updated
2026-11-20
Source record
OAI-PMH GetRecord
related terms
citation

Torabi, Ali. Suicide Ideation Detection using Advanced Signal Processing and Machine Learning Analysis of the EEG. 2026. https://macsphere.mcmaster.ca/handle/11375/32634