Channel Reflection: Knowledge-Driven Data Augmentation for EEG-Based Brain-Computer Interfaces

TL;DR

This work tackles calibration data scarcity in EEG-based BCIs by introducing Channel Reflection (CR), a parameter-free data augmentation that injects prior knowledge of channel distributions through symmetry-based channel exchanges and selective label swapping for left-right motor imagery. CR is demonstrated to be effective, robust, and flexible across eight public EEG datasets spanning four BCIs paradigms (MI, SSVEP, P300, seizures), improving classification performance over Baseline and existing augmentations and compatible with other augmentation methods. Across MI, SSVEP, P300, and seizure tasks, CR consistently enhances accuracy or balanced accuracy and is shown to preserve class structure in augmented data via visualization analyses. The approach is made available with open-source code, and the findings suggest that incorporating prior knowledge into data augmentation should be a standard step in EEG-based BCIs, with future work aimed at extending symmetry assumptions and class coverage.

Abstract

A brain-computer interface (BCI) enables direct communication between the human brain and external devices. Electroencephalography (EEG) based BCIs are currently the most popular for able-bodied users. To increase user-friendliness, usually a small amount of user-specific EEG data are used for calibration, which may not be enough to develop a pure data-driven decoding model. To cope with this typical calibration data shortage challenge in EEG-based BCIs, this paper proposes a parameter-free channel reflection (CR) data augmentation approach that incorporates prior knowledge on the channel distributions of different BCI paradigms in data augmentation. Experiments on eight public EEG datasets across four different BCI paradigms (motor imagery, steady-state visual evoked potential, P300, and seizure classifications) using different decoding algorithms demonstrated that: 1) CR is effective, i.e., it can noticeably improve the classification accuracy; 2) CR is robust, i.e., it consistently outperforms existing data augmentation approaches in the literature; and, 3) CR is flexible, i.e., it can be combined with other data augmentation approaches to further increase the performance. We suggest that data augmentation approaches like CR should be an essential step in EEG-based BCIs. Our code is available online.

Figures (8)

• Figure 1: Knowledge-driven machine learning pipeline for EEG-based BCIs, which includes EEG data acquisition, data preprocessing, data augmentation (optional), feature engineering, and classification/regression. The latter two can be integrated into a single end-to-end neural network. Task-related prior knowledge can be integrated into all blocks of the pipeline.
• Figure 2: CR data augmentation for (a) unipolar channels, using MI-I dataset Tangermann2012BNCI2014001 as an example; and, (b) bipolar channels, using Seizure-I dataset Stevenson2019 as an example.
• Figure 3: Closed-loop EEG-based BCI system, including the proposed CR data augmentation block.
• Figure 4: Illustration of the three experiment scenarios.
• Figure 5: $t$-SNE visualization of CSP features from the original EEG trials and the CR augmented EEG trials. (a) Subjects S0-S8 in MI-I; (b) Subjects S0-S8 in MI-II; and, (c) Subjects S0-S6 in MI-III. Different colors represent different classes. The dots represent the original trials, and the crosses represent the CR augmented trials. Note that for S0 and S5 in MI-III, the classification task was right hand versus feet.