Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sparse Bayesian Modeling of EEG Channel Interactions Improves P300 Brain-Computer Interface Performance

Guoxuan Ma, Yuan Zhong, Moyan Li, Yuxiao Nie, Jian Kang

TL;DR

This work addresses decoding P300 EEG signals in BCI by explicitly modeling interactions across EEG channels within a sparse Bayesian time-varying regression framework. It introduces a relaxed-thresholded Gaussian process prior to induce structured sparsity in both channel effects and inter-channel interactions, enabling automatic temporal feature selection and interpretability. The per-subject SI-RTGP approach yields higher accuracy and BCI-Utility on a 55-participant P300 speller dataset and reveals heterogeneity in which channel interactions are predictive. The results demonstrate that incorporating structured EEG channel interactions improves predictive performance, user-centric throughput, and personalization for P300 BCIs, with practical implications for adaptive neural interfaces.

Abstract

Electroencephalography (EEG)-based P300 brain-computer interfaces (BCIs) enable communication without physical movement by detecting stimulus-evoked neural responses. Accurate and efficient decoding remains challenging due to high dimensionality, temporal dependence, and complex interactions across EEG channels. Most existing approaches treat channels independently or rely on black-box machine learning models, limiting interpretability and personalization. We propose a sparse Bayesian time-varying regression framework that explicitly models pairwise EEG channel interactions while performing automatic temporal feature selection. The model employs a relaxed-thresholded Gaussian process prior to induce structured sparsity in both channel-specific and interaction effects, enabling interpretable identification of task-relevant channels and channel pairs. Applied to a publicly available P300 speller dataset of 55 participants, the proposed method achieves a median character-level accuracy of 100\% using all stimulus sequences and attains the highest overall decoding performance among competing statistical and deep learning approaches. Incorporating channel interactions yields subgroup-specific gains of up to 7\% in character-level accuracy, particularly among participants who abstained from alcohol (up to 18\% improvement). Importantly, the proposed method improves median BCI-Utility by approximately 10\% at its optimal operating point, achieving peak throughput after only seven stimulus sequences. These results demonstrate that explicitly modeling structured EEG channel interactions within a principled Bayesian framework enhances predictive accuracy, improves user-centric throughput, and supports personalization in P300 BCI systems.

Sparse Bayesian Modeling of EEG Channel Interactions Improves P300 Brain-Computer Interface Performance

TL;DR

This work addresses decoding P300 EEG signals in BCI by explicitly modeling interactions across EEG channels within a sparse Bayesian time-varying regression framework. It introduces a relaxed-thresholded Gaussian process prior to induce structured sparsity in both channel effects and inter-channel interactions, enabling automatic temporal feature selection and interpretability. The per-subject SI-RTGP approach yields higher accuracy and BCI-Utility on a 55-participant P300 speller dataset and reveals heterogeneity in which channel interactions are predictive. The results demonstrate that incorporating structured EEG channel interactions improves predictive performance, user-centric throughput, and personalization for P300 BCIs, with practical implications for adaptive neural interfaces.

Abstract

Electroencephalography (EEG)-based P300 brain-computer interfaces (BCIs) enable communication without physical movement by detecting stimulus-evoked neural responses. Accurate and efficient decoding remains challenging due to high dimensionality, temporal dependence, and complex interactions across EEG channels. Most existing approaches treat channels independently or rely on black-box machine learning models, limiting interpretability and personalization. We propose a sparse Bayesian time-varying regression framework that explicitly models pairwise EEG channel interactions while performing automatic temporal feature selection. The model employs a relaxed-thresholded Gaussian process prior to induce structured sparsity in both channel-specific and interaction effects, enabling interpretable identification of task-relevant channels and channel pairs. Applied to a publicly available P300 speller dataset of 55 participants, the proposed method achieves a median character-level accuracy of 100\% using all stimulus sequences and attains the highest overall decoding performance among competing statistical and deep learning approaches. Incorporating channel interactions yields subgroup-specific gains of up to 7\% in character-level accuracy, particularly among participants who abstained from alcohol (up to 18\% improvement). Importantly, the proposed method improves median BCI-Utility by approximately 10\% at its optimal operating point, achieving peak throughput after only seven stimulus sequences. These results demonstrate that explicitly modeling structured EEG channel interactions within a principled Bayesian framework enhances predictive accuracy, improves user-centric throughput, and supports personalization in P300 BCI systems.
Paper Structure (20 sections, 1 theorem, 7 equations, 10 figures, 3 tables)

This paper contains 20 sections, 1 theorem, 7 equations, 10 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Data and motivation
  3. Related work
  4. Our contributions
  5. Method
  6. Bayesian time-varying model with signal interactions
  7. Relaxed-thresholded Gaussian process prior
  8. Posterior Computation
  9. P300 BCI Speller Data Analysis
  10. Benchmark methods and implementation details
  11. Evaluation of prediction performance
  12. Predictive Utility and Heterogeneity of Channel Interactions
  13. When do channel interactions improve prediction performance?
  14. Heterogeneity in channel interaction effects
  15. Temporal dynamics of channel selection
  16. ...and 5 more sections

Key Result

Proposition 1

Given a thresholding parameter $\omega>0$, let $T_r(\theta, \omega, \xi^2) = \theta\cdot I(|\Tilde{\theta}|>\omega)$, $T_h(\theta, \omega) = \theta\cdot I(|{\theta}|>\omega)$ and $T_s(\theta, \omega) = \operatorname{sgn}(\theta)(|\theta| - \omega)\cdot I(|{\theta}|>\omega)$ where $\theta \sim \mathc

Figures (10)

  • Figure 1: An illustration of the P300 BCI spelling task and recorded signal. (A) The P300 BCI speller presents a sequence of row or column flashes (a row or a column) on a virtual screen to the user. The user focuses on a target character and responds to the row and column flashes that contain the target character. The EEG signals of the flash are recorded from multiple channels and are segmented into a 1200 ms window following stimulus onset. (B) The names and layout of the 32 scalp EEG channels according to the international 10-20 system. (C) Example of recorded multi-channel EEG signals and the derived channel connectivity.
  • Figure 2: Illustration of the data and notation for the SI-RTGP model. $\mathbf{X}_{ki}$ represents the EEG signal from channel $k$ for the $i$-th flash. For the $i$-th flash, the signal interaction between channel $k_1$ and channel $k_2$ is measured by the z-transformed Pearson correlation between EEG signals from channels $k_1$ and $k_2$.
  • Figure 3: Illustration of different thresholded Gaussian process prior. $T_s(\cdot, 0.5)$ and $T_h(\cdot, 0.5)$ represents the soft and the hard thresholding function thresholded at 0.5. $T_r(\cdot, 0.5, \xi^2)$ represents the proposed relaxed-thresholding function with different value of relaxing parameter $\xi^2$.
  • Figure 4: Median character prediction accuracy across 55 participants for different methods over 15 sequences using the proposed models (four variants) and competing methods. For clarity, the left panel includes the proposed method with channel interaction and probit link (SITRGP_probit) and competing methods, while the right panel compares the four model variants. Competing methods include GLASS, Logistic Regression (LR), Support Vector Classifier (SVC), Random Forest (RF), Extreme Gradient Boosting (XGBoost), EEGNet, and Stepwise Linear Discriminant Analysis (SWLDA).
  • Figure 5: Median BCI-Utility (bit/second) across 55 participants for different methods over 15 sequences using the proposed models (four variants) and competing methods. For clarity, the left panel includes the proposed method with channel interaction and probit link (SITRGP_probit) and competing methods, while the right panel compares the four model variants. Competing methods include GLASS, Logistic Regression (LR), Support Vector Classifier (SVC), Random Forest (RF), Extreme Gradient Boosting (XGBoost), EEGNet, and Stepwise Linear Discriminant Analysis (SWLDA).
  • ...and 5 more figures

Theorems & Definitions (2)

  • Definition 1
  • Proposition 1