A Local-Global Graph Convolutional Network for Depression Recognition using EEG Signals

Graph Convolutional Networks (GCNs) have shown remarkable capabilities in learning the topological relationships among electroencephalogram (EEG) channels for recognizing depression. However, existing GCN methods often focus on a single spatial pattern, disregarding the relevant connectivity of local functional regions and neglecting the data dependency of the original EEG data. To address these limitations, we introduce the Local-Global GCN (LG-GCN), a novel GCN inspired by brain science research, which learns the local-global graph representation of EEG. Our approach leverages discriminative features extracted from EEG signals as auxiliary information to capture dynamic multi-level spatial information between EEG channels. Specifically, the representation learning of the topological space in brain regions comprises two graphs: one for exploring augmentation information in local functional regions and another for extracting global dynamic information. The aggregation of multiple graphs enables the GCN to acquire more robust features. Additionally, we develop an Information Enhancement Module (IEM) to capture multi-dimensional fused features. Extensive experiments conducted on public datasets demonstrate that our proposed method surpasses state-of-the-art (SOTA) models, achieving an impressive accuracy of 99.30% in depression recognition.

@article{Chen2023,
  title     = {A Local-Global Graph Convolutional Network for Depression Recognition using EEG Signals},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {14},
  number    = {7},
  year      = {2023},
  publisher = {The Science and Information Organization},
  author    = {Yu Chen and Xiuxiu Hu and Lihua Xia},
  doi       = {10.14569/IJACSA.2023.0140720},
  url       = {https://doi.org/10.14569/IJACSA.2023.0140720}
}