Development of machine learning-based connectivity methods for EEG data

This thesis presents a novel approach, the STAGIN-EEG framework, for analyzing brain connectivity using EEG data. STAGIN-EEG adapts the Spatio-Temporal Attention Graph Isomorphism Network (STAGIN) to EEG, exploiting the high temporal resolution of EEG signals to capture dynamic functional connectivity (dFC) patterns. Unlike traditional static functional connectivity methods, the proposed model accounts for the time-varying nature of brain networks, thus providing a more accurate representation of the brain's functional dynamics. This work addresses a critical gap in the literature, where EEG-based functional connectivity has not been extensively explored, especially in the context of dynamic patterns in brain connectivity. The model is applied to the binary classification of abnormal and normal EEG, and the results show promising performance, demonstrating that STAGIN-EEG is a robust tool for dynamic brain connectivity analysis, achieving an accuracy of 78.9% on test set. Findings of this study provide a basis for future research, which can expand on this framework by incorporating additional EEG features and further refining the model to improve its performance and generalizability.