.png){kind=icon}
.png){kind=icon}
Return
Improving depression diagnosis using a brain module-based weighted hypergraph convolutional network framework
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131074.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Traditional functional connectivity network (FCN) is limited to capturing pairwise correlations between electrodes or brain regions. In contrast, hyper-connectivity network (HCN), which can capture complex higher-order relationships among brain regions, has gained increasing attention in brain disease diagnosis. While existing HCN-based models have shown effectiveness in disease recognition, several challenges remain. Reliable HCN construction is hindered by methodological constraints and electroencephalogram (EEG) noise. Moreover, these models often underutilize disease-related abnormal neural mechanisms, resulting in limited interpretability and suboptimal recognition performance. To address these issues, we propose a brain module-based adaptive weighted hypergraph convolutional network (BM-WHGCN) framework to improve depression diagnosis. Firstly, node-level feature representations of EEG are comprehensively extracted from temporal, frequency, and randomness domains. Secondly, a local synchronicity- constrained sparse method is proposed to construct HCN that integrates both global and local correlations. Subsequently, densely connected node subsets are identified as brain modules through modularity optimization, and long-range connections are attenuated to develop a learning framework with enhanced neuropathological interpretability. Finally, an extended hypergraph convolutional network aggregates node features and hyperedges to generate a deep representation of the spatiotemporal information. Extensive experiments conducted on a public and our own EEG dataset demonstrate that BM-WHGCN outperforms other state-of-the-art methods for depression recognition. Additionally, our framework demonstrates strong interpretability by effectively identifying hyperedges and electrode nodes closely associated with depression. In conclusion, the proposed method offers significant potential for the early detection and intervention of depression and can be readily adapted for the diagnosis of other brain disorders.
Keywords:
hyper-connectivity network
depression diagnosis
hypergraph convolutional network
EEG feature extraction
brain modules