P300-Based Machine Learning BCIs for Brain-to-Text Neural Decoding in Paralysed Patients:A Narrative Review

Abstract

This narrative review synthesizes recent advances in brain-to-text neural decoding using P300-based machine learning brain–computer interfaces (BCIs) for individuals with severe motor impairments. It situates the P300 event-related potential within its neurophysiological and theoretical foundations and traces its trajectory from early speller paradigms to contemporary adaptive, hybrid, and deep learning–driven systems. Historical milestones in stimulus presentation, signal processing, and classifier design are examined alongside recent innovations that integrate multimodal ERP features, probabilistic models, and closed-loop neurofeedback to enhance accuracy, robustness, and ecological validity. Clinical applications extend beyond communication restoration to neurorehabilitation, encompassing motor recovery and language training, while real-world deployment increasingly emphasizes portability, cost-effectiveness, and home usability. Critical reflections highlight persistent challenges, including inter-subject variability in P300 morphology, trade-offs between interpretability and accuracy, and underexplored ethical issues related to privacy, agency, and equitable access. The review concludes by framing P300-based brain-to-text BCIs as evolving from static laboratory tools into intelligent, context-aware systems capable of co-adapting with users and shaping the clinical, societal, and economic landscape of assistive neurotechnology.