Design of A Real-Time Motor Imagery BCI-VR System with Sample Segmentation and Soft Voting

Motor imagery-based brain-computer interface integrated with virtual reality (MIBCI-VR) enables hands-free, intuitive interaction in immersive environments. However, systematic online validation of real-time, closed-loop MIBCI-VR systems, especially those employing advanced ensemble decoding strategies, remains limited. This study introduces a novel online MIBCI-VR system that combines EEG sample segmentation and soft voting within a closed-loop framework. Leveraging a 32channel saline electroencephalogram (EEG) device and a VR headset, the system integrates advanced signal preprocessing, deep learning-based EEG Conformer classification, and probabilistic aggregation of segment-level predictions for robust, low-latency control. In real-time VR tasks, participants achieved an average online classification accuracy of 85%, with both prediction and data transmission times averaging 0.12 seconds. Usability evaluations revealed moderate workload, satisfaction, and a strong sense of agency, indicating that users found the system practical and engaging for neuro-interactive tasks. Qualitative feedback further highlighted the intuitive feedback and positive user experience provided by the system. These findings demonstrate both the technical and usability innovation of real-time MIBCI-VR, providing a strong foundation for the next generation of neuro-interactive systems and offering actionable insights for future system optimization and broader adoption.