[Paper Review] PGCN: Pyramidal Graph Convolutional Network for EEG Emotion Recognition

Ming Jin, Enwei Zhu+3 more|arXiv (Cornell University)|Feb 6, 2023

EEG and Brain-Computer Interfaces52 references4 citations

TL;DR

This paper proposes PGCN, a pyramidal graph convolutional network that enhances EEG-based emotion recognition by modeling local, mesoscopic, and global electrode relationships through a three-level architecture. By integrating spatial topology, knowledge-guided brain regions, and position-aware global connectivity, PGCN achieves state-of-the-art performance while mitigating over-smoothing in deep GCNs.

ABSTRACT

Emotion recognition is essential in the diagnosis and rehabilitation of various mental diseases. In the last decade, electroencephalogram (EEG)-based emotion recognition has been intensively investigated due to its prominative accuracy and reliability, and graph convolutional network (GCN) has become a mainstream model to decode emotions from EEG signals. However, the electrode relationship, especially long-range electrode dependencies across the scalp, may be underutilized by GCNs, although such relationships have been proven to be important in emotion recognition. The small receptive field makes shallow GCNs only aggregate local nodes. On the other hand, stacking too many layers leads to over-smoothing. To solve these problems, we propose the pyramidal graph convolutional network (PGCN), which aggregates features at three levels: local, mesoscopic, and global. First, we construct a vanilla GCN based on the 3D topological relationships of electrodes, which is used to integrate two-order local features; Second, we construct several mesoscopic brain regions based on priori knowledge and employ mesoscopic attention to sequentially calculate the virtual mesoscopic centers to focus on the functional connections of mesoscopic brain regions; Finally, we fuse the node features and their 3D positions to construct a numerical relationship adjacency matrix to integrate structural and functional connections from the global perspective. Experimental results on three public datasets indicate that PGCN enhances the relationship modelling across the scalp and achieves state-of-the-art performance in both subject-dependent and subject-independent scenarios. Meanwhile, PGCN makes an effective trade-off between enhancing network depth and receptive fields while suppressing the ensuing over-smoothing. Our codes are publicly accessible at https://github.com/Jinminbox/PGCN.

Motivation & Objective

• Address the limitation of shallow GCNs in capturing long-range electrode dependencies in EEG-based emotion recognition.
• Overcome the over-smoothing problem in deep GCNs while expanding receptive fields for better modeling of functional and structural brain connectivity.
• Integrate prior neuroanatomical knowledge into GCN architecture to improve representation of mesoscopic brain regions.
• Balance feature aggregation across local, mesoscopic, and global scales to enhance emotion recognition performance on both subject-dependent and subject-independent benchmarks.
• Develop a scalable, publicly available framework for EEG emotion recognition using multi-level graph learning.

Proposed method

• Construct a vanilla GCN based on 3D topological relationships among EEG electrodes to extract second-order local features.
• Define mesoscopic brain regions using prior anatomical knowledge (e.g., frontal, temporal, occipital lobes) and apply mesoscopic attention to compute virtual centers that emphasize functional connectivity within these regions.
• Create a position-aware numerical adjacency matrix by fusing node features with their 3D spatial coordinates to encode structural and functional connectivity at the global level.
• Integrate features from all three levels—local, mesoscopic, and global—through a hierarchical fusion strategy to enrich representations across scales.
• Use a learnable adjacency matrix in the global module to capture sparse, long-range dependencies while minimizing over-smoothing.
• Train the end-to-end PGCN model using cross-entropy loss with stochastic gradient descent, optimizing for emotion classification on EEG data.

Experimental results

Research questions

• RQ1 Can a multi-scale graph convolutional architecture effectively model local, mesoscopic, and global brain connectivity in EEG-based emotion recognition?
• RQ2 How does integrating prior neuroanatomical knowledge into GCN architecture improve performance on EEG emotion recognition tasks?
• RQ3 To what extent can a pyramidal GCN design extend the receptive field without inducing over-smoothing in deep networks?
• RQ4 Does the proposed method achieve superior performance compared to existing GCN-based models in both subject-dependent and subject-independent emotion recognition settings?
• RQ5 How do learned attention weights and adjacency matrices reflect known functional brain networks related to emotion processing?

Key findings

• PGCN achieves state-of-the-art performance on three public EEG emotion recognition datasets: SEED, SEED-IV, and DEAP.
• The model significantly improves classification accuracy in both subject-dependent and subject-independent settings, outperforming existing GCN-based methods.
• Visualization of learned adjacency matrices shows that electrodes in the frontal, temporal, and occipital lobes consistently receive higher self-weighting and form critical connections, aligning with known emotion-related brain regions.
• Top-10 connections in the learned adjacency matrix are predominantly located in the frontal, temporal, and occipital lobes, indicating their functional importance in emotion processing.
• The performance curve shows that accuracy increases with depth up to a point, after which over-smoothing dominates, confirming the effectiveness of PGCN’s design in delaying this degradation.
• The proposed method effectively balances expanded receptive fields and over-smoothing, demonstrating stable performance even with deeper architectures.