Towards a Fast Steady-State Visual Evoked Potentials (SSVEP) Brain-Computer Interface (BCI)

Aung Aung Phyo Wai, Yangsong Zhang, Heng Guo, Ying Chi, Lei Zhang, Xian-Sheng Hua, Seong-Whan Lee, Cuntai GuanarXiv 2020 · OpenReview

TL;DR

This paper introduces CSTA, a novel training-free method for Steady-State Visual Evoked Potentials (SSVEP) Brain-Computer Interfaces (BCIs) that achieves high accuracy in sub-second response times. By combining spatial filtering and temporal alignment, CSTA significantly improves SSVEP recognition performance, making BCIs more practical and user-friendly.

What Problem Does This Paper Address?

Brain-Computer Interfaces (BCIs) offer a way for individuals to interact with technology using their brain signals. SSVEP-based BCIs are particularly promising due to their reliability and high information throughput. However, achieving high accuracy with SSVEP typically requires long recording times (one second or more), which can be cumbersome and lead to user discomfort. While various methods have been proposed to improve accuracy in shorter time windows, they often rely on extensive subject-specific training and calibration, which is time-consuming and can be a barrier to usability. This paper addresses the need for a BCI system that can achieve high accuracy in very short response times (sub-second) without requiring any user-specific training or calibration.

Methods

The core of the proposed method, CSTA (Combined Spatial Filtering and Temporal Alignment), lies in its two-pronged approach:

  • CCA Spatial Filtering: Canonical Correlation Analysis (CCA) is used to find the optimal linear relationship between multi-channel EEG signals and reference signals corresponding to different stimulus frequencies. This step extracts spatial features that maximize the correlation, effectively filtering out noise and enhancing the relevant SSVEP signals.
  • DTW Temporal Alignment: Dynamic Time Warping (DTW) is employed to align the temporal patterns of the EEG signals with stimulus templates. DTW is adept at handling non-linear temporal distortions and phase shifts, allowing for robust matching even when the exact timing of the brain response deviates slightly from the stimulus.

CSTA combines these two techniques by first using CCA to extract spatially filtered EEG signals. These filtered signals are then fed into the DTW module. Furthermore, CSTA utilizes multi-phased templates in the DTW step, which helps to account for potential phase variations in the SSVEP responses. The final decision is made by fusing the outputs from both the CCA (correlation vector) and DTW (alignment vector) components. This complementary fusion aims to leverage the strengths of both linear correlation and non-linear temporal matching.

Datasets

The study evaluated CSTA using two publicly available SSVEP datasets:

  • OpenBMI Dataset: This dataset features a 4-class SSVEP interface with distinct stimulus frequencies. It was collected by Korea University.
  • HS-SSVEP Dataset: This dataset is a benchmark dataset from Tsinghua University, featuring a 40-class SSVEP interface with unique frequency and phase information for each stimulus.

The authors also compared CSTA against three state-of-the-art (SOTA) methods: FBCCA (a training-free method), TRCA (a training-based method), and IT-CCA (another training-based method).

Key Insights and Ideas

  • The Power of Fusion: Combining linear spatial filtering (CCA) with non-linear temporal alignment (DTW) proves to be a potent strategy for SSVEP recognition, especially in short time windows.
  • Training-Free Advantage: The proposed CSTA method demonstrates that high accuracy can be achieved without the need for subject-specific training, significantly reducing user burden and increasing practical usability.
  • Sub-Second Performance is Achievable: The results strongly suggest that SSVEP BCIs can achieve reliable performance in response times as short as 0.2-0.5 seconds, opening doors for more responsive and real-time applications.
  • Robustness Across Datasets: CSTA's consistent performance across two datasets with different characteristics (number of classes, stimulus coding) highlights its generalizability.

Key Results

  • High Accuracy in Short Windows: CSTA achieved a maximum mean accuracy of 97.43 ± 2.26% on the 4-class OpenBMI dataset and 85.71 ± 13.41% on the 40-class HS-SSVEP dataset, all within sub-second response times.
  • Significant Improvement Over Training-Free Methods: CSTA yielded significantly higher mean performance (p < 0.001) compared to the FBCCA training-free method on both datasets.
  • Competitive with Training-Based Methods: Compared to training-based methods (TRCA and IT-CCA), CSTA showed a 29.33 ± 19.65% higher mean accuracy with statistically significant differences in time windows less than 0.5 seconds. In longer time windows, CSTA performed comparably or better.
  • Lower Variability: CSTA exhibited less variability in mean accuracy and Information Transfer Rate (ITR) across different time windows compared to the other methods.

Limitations and Future Work

  • Empirical Specification of Phases: The authors acknowledge that a limitation of CSTA is the need to empirically specify the number of phases (k) used in the multi-phased templates for temporal alignment. They raise the question of how to systematically optimize this parameter.
  • Subject-Independent Evaluation: While CSTA is designed to be subject-independent, the authors note that most training-based methods are evaluated using subject-specific cross-validation. They suggest that further evaluation using subject-independent leave-one-subject-out scenarios would be beneficial.
  • Deep Learning Comparison: The paper mentions that deep learning approaches are increasingly used in BCI research and suggests future work could involve evaluating CSTA against deep learning-based SSVEP recognition methods.
  • Improving Temporal Alignment: Future work could also explore improved time-series alignment mapping methods to further enhance CSTA's performance.

Summary updated

Abstract

Steady-state visual evoked potentials (SSVEP) brain-computer interface (BCI) provides reliable responses leading to high accuracy and information throughput. But achieving high accuracy typically requires a relatively long time window of one second or more. Various methods were proposed to improve sub-second response accuracy through subject-specific training and calibration. Substantial performance improvements were achieved with tedious calibration and subject-specific training; resulting in the user's discomfort. So, we propose a training-free method by combining spatial-filtering and temporal alignment (CSTA) to recognize SSVEP responses in sub-second response time. CSTA exploits linear correlation and non-linear similarity between steady-state responses and stimulus templates with complementary fusion to achieve desirable performance improvements. We evaluated the performance of CSTA in terms of accuracy and Information Transfer Rate (ITR) in comparison with both training-based and training-free methods using two SSVEP data-sets. We observed that CSTA achieves the maximum mean accuracy of 97.43$\pm$2.26 % and 85.71$\pm$13.41 % with four-class and forty-class SSVEP data-sets respectively in sub-second response time in offline analysis. CSTA yields significantly higher mean performance (p<0.001) than the training-free method on both data-sets. Compared with training-based methods, CSTA shows 29.33$\pm$19.65 % higher mean accuracy with statistically significant differences in time window less than 0.5 s. In longer time windows, CSTA exhibits either better or comparable performance though not statistically significantly better than training-based methods. We show that the proposed method brings advantages of subject-independent SSVEP classification without requiring training while enabling high target recognition performance in sub-second response time.
BibTeX
@misc{wai2020towards,
  author = {Aung Aung Phyo Wai and Yangsong Zhang and Heng Guo and Ying Chi and Lei Zhang and Xian-Sheng Hua and Seong-Whan Lee and Cuntai Guan},
  title = {{Towards a Fast Steady-State Visual Evoked Potentials (SSVEP) Brain-Computer Interface (BCI)}},
  year = {2020},
  eprint = {2002.01171},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2002.01171},
}
Figure