Pseudo Channel: Time Embedding for Motor Imagery Decoding

TL;DR

This paper addresses the challenge of decoding motor-imagery EEG with strong inter-subject variability by introducing traveling-wave based time embedding as a pseudo channel. The method encodes temporal dynamics via a traveling-wave formulation and is compared against Transformer-style position encoding across ConvNet and EEGNet architectures on two MI-EEG datasets (BCI4-2A and BCI4-2B). Results show that traveling-wave time embedding generally improves decoding accuracy and robustness, particularly for participants with previously hard-to-decode signals, and often outperforms position encoding. The work offers a new direction for EEG decoding and attention-mechanism research, with implications for improved MI-BCI performance and deeper insights into temporal dynamics of EEG signals, while acknowledging limitations like a single-peak model and limited hyperparameter exploration.

Abstract

Motor imagery (MI) based EEG represents a frontier in enabling direct neural control of external devices and advancing neural rehabilitation. This study introduces a novel time embedding technique, termed traveling-wave based time embedding, utilized as a pseudo channel to enhance the decoding accuracy of MI-EEG signals across various neural network architectures. Unlike traditional neural network methods that fail to account for the temporal dynamics in MI-EEG in individual difference, our approach captures time-related changes for different participants based on a priori knowledge. Through extensive experimentation with multiple participants, we demonstrate that this method not only improves classification accuracy but also exhibits greater adaptability to individual differences compared to position encoding used in Transformer architecture. Significantly, our results reveal that traveling-wave based time embedding crucially enhances decoding accuracy, particularly for participants typically considered "EEG-illiteracy". As a novel direction in EEG research, the traveling-wave based time embedding not only offers fresh insights for neural network decoding strategies but also expands new avenues for research into attention mechanisms in neuroscience and a deeper understanding of EEG signals.

Figures (5)

• Figure 1: Normalized energy distribution of EEG signals across different time points and frequency bands during a single trial for various participants. The x-axis represents the time during the imagery period, with time zero marking the commencement of motor imagery. The energy in the time-frequency plots is normalized.
• Figure 2: Visualization of Various Time-Embedding Results and Their Integration into EEG Signals. Subplot (a) illustrates the waveform variations of the traveling-wave based time embedding across different values of $\lambda$. Subplot (b) displays the waveform changes of the traveling-wave based time embedding with varying $t$ values. Subplot (c) uses a scatter plot to depict position encoding, highlighting its non-continuous distribution with each position uniquely encoded. Subplot (d) demonstrates the integration of time embedding as a pseudo channel within the EEG.
• Figure 3: Typical EEG data collection process during motor imagery trials, illustrating the use of EEG signals acquired during the motor imagery period for decoding.
• Figure 4: Variations in decoding accuracy across participants under different hyperparameters of traveling-wave based time embedding compared to benchmark networks. Positive values on the vertical axis indicate higher decoding accuracy compared to the benchmark network, whereas negative values signify lower decoding accuracy than the benchmark network.
• Figure 5: Average accuracy variations (across all participants) by parameter combinations in traveling-wave based time embedding.Positive values on the vertical axis indicate higher decoding accuracy compared to the benchmark network, whereas negative values signify lower decoding accuracy than the benchmark network.