The MindGomoku: An Online P300 BCI Game Based on Bayesian Deep Learning

Area of Science:

• Neuroscience
• Human-Computer Interaction
• Game Design

Background:

• Brain-computer interface (BCI) technology offers potential beyond assistive devices, extending into entertainment.
• Current BCI games face limitations in control performance and user fatigue, hindering widespread adoption.
• Developing practical and engaging BCI gaming experiences requires integrating game design with BCI system capabilities.

Purpose of the Study:

• To propose and evaluate MindGomoku, a P300 BCI game designed for natural and feasible gameplay using electroencephalogram (EEG) signals.
• To demonstrate a novel approach by integrating game rules with BCI system design for enhanced user experience.
• To introduce a simplified Bayesian convolutional neural network (SBCNN) algorithm for accurate BCI control with limited training data.

Main Methods:

• Development of the MindGomoku game, incorporating P300 BCI paradigms tailored to game mechanics.
• Implementation of a simplified Bayesian convolutional neural network (SBCNN) algorithm for processing EEG signals.
• Conducting online control experiments with 10 subjects to assess system reliability and performance.

Main Results:

• All 10 subjects successfully controlled the MindGomoku game.
• The system achieved an average control accuracy of 90.7%.
• Subjects played the game for an average of over 11 minutes, demonstrating sustained engagement.

Conclusions:

• The proposed MindGomoku system demonstrates stability and effectiveness for BCI-based gaming.
• The SBCNN algorithm enables high accuracy with limited EEG training samples.
• This research expands the potential of BCI technology for entertainment and accessibility, particularly for individuals with disabilities.