EEG motor imagery decoding: A framework for comparative analysis with channel attention mechanisms

TL;DR

This paper tackles EEG motor imagery decoding by introducing a simple, lightweight BaseNet baseline and a modular framework to evaluate a broad set of channel attention mechanisms under fair, fixed training and data conditions. The methodology spans four diverse datasets and two evaluation settings (within-subject and cross-subject), enabling robust cross-dataset comparisons. Results show that BaseNet is highly competitive with state-of-the-art, while channel attention modules offer modest gains that depend on dataset and sensor count; larger, more complex models may outperform in specific settings but at substantial computational cost. Overall, the framework provides a practical foundation for rapid, fair comparison of attention mechanisms in BCIs, with strong generalization across datasets and clear guidance on when attention adds value for real-time EEG decoding.

Abstract

The objective of this study is to investigate the application of various channel attention mechanisms within the domain of brain-computer interface (BCI) for motor imagery decoding. Channel attention mechanisms can be seen as a powerful evolution of spatial filters traditionally used for motor imagery decoding. This study systematically compares such mechanisms by integrating them into a lightweight architecture framework to evaluate their impact. We carefully construct a straightforward and lightweight baseline architecture designed to seamlessly integrate different channel attention mechanisms. This approach is contrary to previous works which only investigate one attention mechanism and usually build a very complex, sometimes nested architecture. Our framework allows us to evaluate and compare the impact of different attention mechanisms under the same circumstances. The easy integration of different channel attention mechanisms as well as the low computational complexity enables us to conduct a wide range of experiments on four datasets to thoroughly assess the effectiveness of the baseline model and the attention mechanisms. Our experiments demonstrate the strength and generalizability of our architecture framework as well as how channel attention mechanisms can improve the performance while maintaining the small memory footprint and low computational complexity of our baseline architecture. Our architecture emphasizes simplicity, offering easy integration of channel attention mechanisms, while maintaining a high degree of generalizability across datasets, making it a versatile and efficient solution for EEG motor imagery decoding within brain-computer interfaces.

Figures (8)

• Figure 1: Channel attention mechanisms. GAP = global average pooling, FC = fully-connected layer, Cov=covariance pooling, BN=batch normalization, RW Conv = row-wise convolution, $\text{DCT}_\text{group}$=grouped discrete cosine transform, Enc=encoder block, CN=channel normalization, STD = global standard deviation, CFC= channel-wise fully-connected layer. Adapted from guo2022attentionreview.
• Figure 2: Architecture and development of BaseNet. (a): g = number of groups, BN = batch normalization. The red dot indicates the position of the optional channel attention mechanism. (b): Accuracies for the BCIC IV 2a dataset in the within-subject scenario. The dashed bar represents the reported accuracy of ShallowNet with their cropped training strategy. The black error bars indicate the standard deviation for five runs with different random seeds.
• Figure 3: Ablation studies investigating the influence of the reduction rate and the kernel size for SENet and ECANet respectively. The black lines indicate the average test accuracy of BaseNet.
• Figure 4: Ablation studies investigating the relationship between the reduction rate $r$ in the channel attention module and the kernel size $k$ in the temporal attention module of CBAM. The best configuration is indicated by a red frame.
• Figure 5: Ablation studies investigating the relationship between the reduction rate $r$ in the channel attention module and the kernel size $k$ in the temporal attention module of CAT. The best configuration is indicated by a red frame.