Abstracts

Rationale: A brain-computer interface (BCI) is a system that allows individuals with severe neuromuscular disorders to communicate and control devices using their brain waves. Over two million people in the USA may benefit from assistive devices controlled by a BCI. Disabled subjects have used scalp EEG-based BCI paradigms to reliably control personal computers. Electrocorticography (ECoG) has also recently been demonstrated to be a viable control for a BCI. The current EEG signals used to operate our visual evoked BCI have not been characterized in ECoG. We hypothesize that ECoG-translated BCI control signals will provide superior speed and accuracy over scalp-recorded EEG for the symbol selection matrix paradigm. Methods: Subjects: Two patients with medically refractory epilepsy undergoing Phase 2 intracranial grid and depth electrode monitoring for seizure localization were studied. Data Acquisition and Task: Concurrent with the clinical ECOG recording, a 16-channel subset of each patient’s electrodes was monitored using a separate EEG amplification system and BCI2000 software. Each patient sat approximately 75 cm from a video screen with a 6 X 6 square matrix of alphanumeric characters displayed. The task was to focus attention on a prescribed character from the matrix and silently count the number of times the prescribed character randomly flashes until a new character is specified for selection. All data was collected in the copy mode: a character string is presented on the top left of the video monitor and the current prescribed character is highlighted at the end of the character string. A single session was conducted without feedback to the user. The session consisted of eight experimental runs; each run is composed of a series of characters that form a word as chosen by the investigator. There were 32 character epochs total in a session. Data Analysis: For each patient, data from the first four runs were preprocessed and used to train a linear classifier using Stepwise Linear Discriminant Analysis (SWLDA). This classifier was tested using the four subsequent runs to determine the classifier’s ability to predict the intended target from independent data. Results: This is a preliminary analysis of an ongoing study. For both patients, the classifier was able to predict the intended target character at 100% accuracy. The minimum number of sequential stimulus responses that were averaged in order to achieve 100% classification accuracy for each patient were 11 and 6, corresponding to approximately 39 and 23 seconds per character selection, respectively. However, assessing the speed vs. accuracy of the classifier, a superior communication rate can potentially be achieved for the first patient using only 5 flashes, corresponding to approximately 20 seconds per character selection. Conclusions: These preliminary results indicate that a superior BCI communication rate can be achieved using ECoG signals compared to scalp-recorded EEG signals. Further improvements may be realized by systematic characterization of the ECoG responses.