Graph convolutional network-based harmonization of EEG for cross-dataset transfer in motor imagery in BCI

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Motor Imagery Brain-Computer Interface (MI-BCI) systems face challenges due to varying electroencephalogram (EEG) electrode configurations across datasets.
• This heterogeneity limits transfer learning and overall system performance, especially with small datasets.

Purpose of the Study:

• To propose a spatial harmonization framework for mapping heterogeneous EEG recordings to a common physical electrode montage.
• To preserve task-relevant motor imagery information during harmonization.
• To enhance transfer learning and system performance in MI-BCI applications.

Main Methods:

• Modeled each EEG trial as a graph with electrodes as nodes and samples as embeddings.
• Utilized a two-layer Graph Convolutional Network (GCN) to capture spatio-temporal relationships.
• Evaluated harmonization using time, frequency, and spatial domain analyses, and downstream MI classification with EEGNet, FBCNet, and ADFCNN across three public datasets.

Main Results:

• The GCN framework achieved lower harmonization error than spherical spline interpolation.
• Harmonized EEG improved within-dataset classification accuracy, e.g., from 56.57% to 66.20% (EEGNet) and 61.96% to 72.54% (FBCNet).
• The harmonized representation facilitated both source-only cross-dataset transfer and fine-tuning, with combined approaches yielding superior performance.

Conclusions:

• The proposed method effectively addresses electrode-layout incompatibility at the EEG signal level.
• It enables heterogeneous MI EEG datasets to be represented in a common physical montage without restricting to shared electrodes.
• This provides a practical foundation for signal-level harmonization, dataset augmentation, and cross-dataset MI decoding.