ChatBCI-Assist: An Intent-Based P300 Speller With A Locally Deployed LLM and Adaptive Stopping Strategy Enabling...

P300-based speller brain computer interfaces (BCIs) provide promising communication solutions for individuals with severe motor impairments, such as those with amyotrophic lateral sclerosis (ALS). However, existing P300 spellers are constrained by slow typing speed and limited efficiency. Here, we present ChatBCI-Assist, an intent-based P300 speller that integrates a locally deployed large language model (LLM), fine-tuned for the task at hand, and a graphical user interface (GUI) designed for efficient message composition, while incorporating an adaptive stopping strategy for key selection, to achieve record online spelling performance. The LLM, trained on an ALS-specific communication corpus using low-rank adaptation (LoRA), produces context-aware, prefix-constrained word and phrase predictions in real time. The proposed GUI enables efficient, user-adaptive interaction with these predictions and facilitates semantically driven message composition, where suggestions can reflect the user's intended message. To evaluate intent-based communication, we introduce semantic spelling tasks (beyond traditional copy-spelling), where users convey intended meaning rather than reproduce text verbatim, along with metrics to assess communication based on semantic similarity. Results from online experiments show that ChatBCI-Assist achieves record performance, with an average (estimated) information transfer rate (ITR) of 105.2 bits/min, an overall character-level mutual information rate (MIR) of 52.9 bits/min, and average values of 19.7 characters per minute (CPM) in copy-spelling tasks and 30.7 CPM in semantic spelling tasks. Evaluated using semantic ITR (SITR), a metric proposed to characterize semantic communication efficiency, ChatBCI-Assist achieves an SITR of 147.1 bits/min. User experience evaluations further indicate reduced workload and improved usability compared to traditional copy-spelling paradigms. These results demonstrate that integrating locally-adapted LLMs with intent-driven design and subject-specific decoding optimization can substantially improve the speed, efficiency, and user experience of BCI-based communication systems.