A Brain Switch for SSVEP-Based BCI Speller Using an RNN-Based Detection Approach.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Steady-state visual evoked potentials (SSVEP) are common in brain-computer interface (BCI) spellers due to high accuracy and information transfer rates.
• Asynchronous BCIs offer greater practicality but require robust intention detection mechanisms, unlike synchronous systems.
• Existing SSVEP paradigms exhibit frequency design variability, necessitating improved intention recognition.

Purpose of the Study:

• To develop and evaluate a robust two-stage asynchronous BCI system for reliable user intention detection.
• To combine autocorrelation and Long Short-Term Memory (LSTM) for intention detection with an EEGNet classifier.
• To address the need for dependable intention distinction in asynchronous BCI systems.

Main Methods:

• A novel two-stage asynchronous BCI system was proposed.
• The system integrated a robust brain switch model utilizing autocorrelation and Long Short-Term Memory (LSTM) for intention detection.
• An EEGNet-based classifier was employed for signal processing and classification.

Main Results:

• The system demonstrated high detection performance with 98.24 ± 2.21% sensitivity and 82.28 ± 11.63% specificity using 1-second epochs.
• Classification accuracy reached 77.05 ± 14.95% on a 40-class SSVEP dataset from 40 subjects.
• The proposed model showed significant potential for developing more realistic and practical asynchronous BCI systems.

Conclusions:

• The developed two-stage asynchronous BCI system effectively distinguishes user intentions with high accuracy and reliability.
• The integration of autocorrelation, LSTM, and EEGNet provides a robust framework for advanced BCI applications.
• This research contributes to the advancement of practical and user-friendly asynchronous BCI systems.