The P300 speller is a well-established brain-computer interface (BCI) designed to restore communication ability to patients with severe neuromuscular disorders. Traditionally, this system is implemented using scalp electroencephalography (EEG) to detect evoked responses. Using electrocorticography (ECoG) could significantly improve performance due to its superior signal-to-noise ratio.

     The goal of the study is to show that using ECoG in this system can provide a significant benefit over the EEG approach.

       Two patients were studied.

     We conducted this study using two epilepsy patients with implanted ECoG grids in the temporal and occipital lobes. Each patient was presented a grid of characters and told to focus in a specific letter while ECoG data was recorded. Groups of characters were then intensified randomly and we used BCI2000 to detect electrophysiological responses to determine the target character. Additionally, natural language processing (NLP) algorithms were used to utilize domain knowledge to further improve classification accuracy.

     The subject with an occipital grid achieved a typing speed of 6.7 selections/minute with 100% accuracy, resulting in a bit rate of 34.9. When NLP and dynamic trial length were implemented into the system, the bit rate increased further to 43.7. The subject with the temporal grid achieved a bit rate of 18.7, which improved to 21.4 with the implementation of NLP.

     This is a retrospective study.

     The results from this ECoG-based system were significantly better than those from EEG-based systems suggesting that it may be beneficial to implant subdural electrodes in some severely disabled patients.

     While temporal electrodes were sufficient for BCI communication, occipital electrodes provided significantly better results.