Reliable predictor of BCI motor imagery performance using median nerve stimulation

Valérie Marissens Cueva, Laurent Bougrain, Fabien Lotte, Sébastien RimbertJournal of Neural Engineering 2025 · DOI

TL;DR

This paper introduces a novel, non-invasive method using median nerve stimulation (MNS) to predict a user's performance in motor imagery-based Brain-Computer Interfaces (BCIs). By analyzing specific brainwave patterns (ERD) evoked by MNS, researchers can accurately estimate how well a person will perform on a BCI task, potentially streamlining BCI use and improving rehabilitation outcomes.

What Problem Does This Paper Address?

Brain-Computer Interfaces (BCIs) that rely on motor imagery (MI) – imagining a movement – hold great promise for communication, control, and rehabilitation. However, a significant challenge is the variability in user performance, with many individuals struggling to achieve reliable control (often termed "BCI illiteracy"). This variability makes it difficult to optimize BCI systems, personalize training, and effectively use BCIs in clinical settings. Currently, predicting a user's potential BCI performance before extensive training or testing is difficult. This paper aims to bridge this gap by identifying a reliable, non-invasive predictor of MI-BCI performance.

Methods

The core of this study is the use of Median Nerve Stimulation (MNS) as a proxy for motor imagery. MNS, applied to the wrist, elicits brainwave patterns, specifically Event-Related Desynchronization (ERD), in the sensorimotor cortex that are similar to those generated during MI tasks. The researchers hypothesized that the strength of this MNS-induced ERD could predict a user's subsequent MI-BCI performance.

Key methodological steps include:

  • MNS-induced ERD/ERS Quantification: EEG signals were recorded during MNS. The ERD/ERS percentage was calculated within specific frequency bands (mu, beta, mu+beta) and time windows (200-800 ms post-stimulation).
  • Offline MI-BCI Performance Calculation: Standard BCI classification (MI vs. rest) was performed using a Riemannian classifier (Tangent Space Logistic Regression) on the same EEG data, but during actual motor imagery tasks.
  • Correlation and Prediction: Spearman's correlation was used to assess the relationship between MNS-induced ERD and MI-BCI performance. A Least Absolute Shrinkage and Selection Operator (LASSO) regression model was employed to predict MI-BCI performance based on MNS features.
  • Leave-One-Subject-Out Cross-Validation: This rigorous validation method was used to ensure the predictor's generalizability across different users.

Datasets

The study utilized two datasets comprising a total of 31 right-handed healthy subjects. These datasets were originally collected for studies investigating general anesthetic conditions on EEG, but the researchers focused on participants not administered general anesthesia. The data was collected using a 128-channel EEG system.

Key Insights and Ideas

  • MNS as a "Motor Readiness" Indicator: The study strongly suggests that the brain's response to MNS (specifically ERD) reflects a user's underlying capacity for motor imagery, acting as a non-invasive measure of "motor readiness."
  • Predictive Power Beyond Correlation: The research goes beyond simply showing a correlation; it demonstrates that MNS features can predict future MI-BCI performance, a crucial step for practical application.
  • Importance of Frequency Bands and Electrode Location: The mu+beta frequency band and the C3 electrode (with a Laplacian filter) emerged as particularly informative for both predicting and correlating MI-BCI performance.
  • Efficiency of Prediction: The study highlights that a reliable predictor can be established with a relatively short MNS stimulation period (as little as 1-4 minutes), significantly reducing the time and effort required for BCI assessment.
  • LASSO for Feature Selection: The use of LASSO regression effectively identified the most relevant MNS-derived features for prediction, enhancing interpretability and robustness.

Key Results

  • Strong Correlation: The minimum ERD induced by MNS showed a robust negative correlation with MI-BCI performance accuracy, with the strongest correlation (rho = -0.71, p < 0.001) observed in the mu+beta band using the C3 Laplacian.
  • Accurate Prediction: The LASSO model successfully predicted MI-BCI performance with a significant correlation (rho = 0.61, mean absolute error = 9.0, p < 0.01).
  • Time Efficiency: A significant correlation could be achieved with as few as 10 MNS stimulations (less than a minute of data acquisition), and optimal results were obtained with around 40 stimulations (less than 4 minutes).
  • Frequency Band Performance: The mu+beta frequency band generally yielded the best results for both correlation and prediction.

Limitations and Future Work

  • Eyes Closed Condition: The study was conducted with participants' eyes closed for both MI and MNS tasks. Future work should validate these findings with eyes-open conditions, as BCIs are often used in this state.
  • Right-Handed Dominance: All participants were right-handed, and only the right median nerve was stimulated. Future research should investigate the generalizability to left-handed individuals and explore stimulation of the left median nerve.
  • MI vs. Rest vs. Multiple MI Classes: The current predictor focuses on MI vs. rest performance. Future work could explore its utility in predicting performance for tasks involving multiple MI classes (e.g., left vs. right hand).
  • Individualization: While the predictor is user-independent in its application, further personalization by selecting optimal frequency bands, time windows, and electrodes for each individual is suggested.
  • Integration with Other Predictors: Combining MNS-based features with other predictors (e.g., spectral entropy, resting-state mu peak) could further enhance prediction accuracy.
  • BCI Illiteracy Identification: The authors suggest isolating users with the lowest BCI performance to specifically assess the predictor's ability to identify BCI illiteracy.

Summary updated

Abstract

<i>Objective.</i>Predicting performance in brain-computer interfaces (BCIs) is crucial for enhancing user experience, optimizing training and identifying the most efficient BCI approach for each individual.<i>Approach.</i>This study explores the use of median nerve stimulation (MNS) as a predictor of motor imagery (MI)-BCI performance. MNS induces event related (de)synchronization (ERD/ERS) patterns in the brain that are similar to those generated during MI tasks, providing a non-invasive, user-independent, and easy-to-setup method for performance prediction.<i>Main results.</i>Our proposed predictor, based on the minimum value of the ERD induced by the MNS, not only exhibits a robust correlation with the MI-BCI performance accuracy (rho = -0.71,p<0.001), but also effectively predicts this performance with a significant correlation (rho = 0.61, mean absolute error = 9.0,p<0.01). These results demonstrate its validity as a reliable predictor of MI-BCI performance.<i>Significance.</i>By systematically analyzing patterns induced by MNS and correlating them with subsequent MI-BCI task performance, we aim to establish a robust predictive method of motor activity to each individual only based on MNS, making it possible, among other things, to passively predict BCI deficiency or proficiency, and to potentially adapt BCI parameters for an efficient BCI experience or BCI-based recovery.
BibTeX
@article{cueva2025reliable,
  author = {Valérie Marissens Cueva and Laurent Bougrain and Fabien Lotte and Sébastien Rimbert},
  title = {{Reliable predictor of BCI motor imagery performance using median nerve stimulation}},
  journal = {Journal of Neural Engineering},
  year = {2025},
  doi = {10.1088/1741-2552/adc48d},
}
Figure