Cross-Subject P300-Based Audiovisual BCI System via Continual Learning: A Clinical Application for Disorders of Consciousness

This study proposes an advanced cross-subject P300-based audiovisual brain-computer interface (BCI) system to assess consciousness levels and predict clinical outcomes in patients with disorders of consciousness (DOC). The system employs an audiovisual stimulus paradigm, integrating face photos and corresponding name sounds, to enhance the elicitation of P300 signals. It further incorporates a hybrid prototype-based continual learning method (HPC) to improve diagnostic accuracy and robustness. The HPC constructs P300 prototypes for each historical task and selectively integrates both similar and dissimilar prototypes when a new task is introduced. Dissimilar prototypes are hybridized and masked, while similar prototypes are merged via an attention mechanism, effectively preventing catastrophic forgetting. Experimental results demonstrate the efficacy of this approach, with the HPC achieving 98.33% accuracy in a P300 spelling task among healthy subjects and 95.50% accuracy in healthy controls within a clinical setting. Significantly, eight out of ten DOC patients exhibited notable accuracy, underscoring the system's clinical potential. This BCI system thus offers a robust and adaptable solution for assessing consciousness levels and predicting outcomes in DOC patients, contributing to enhanced clinical diagnosis and prognosis.