Event-related potentials from the human cortex as the basis of a direct brain interface for the operation of assistive technology.

A direct brain interface can be defined as a human-computer interface that accepts voluntary commands directly from the brain. Such an interface could be used by people with disabilities to control assistive technology and improve their independence and quality of life. In this dissertation a method for identifying event-related potentials (ERPs) corresponding to voluntary actions is described as a means of developing a direct brain interface. The subjects for this research were patients in an epilepsy surgery program who had cortical electrodes placed as part of their treatment. The electrocorticogram (ECoG) at one electrode location for a particular subject, task, and session was considered as one channel. A total of 1710 channels were obtained. The average ECoG related to a task was calculated for each channel by triggered averaging. ERPs were identified on 164 (9.6%) of the ECoG channels through statistical comparison of a pre-activity and a peak signal interval. Cross-correlation was performed between averaged ERP templates from the first half of a channel and ECoG from the second half. Detections (defined when the cross-correlation values exceeded an experimentally determined threshold) that occurred within 1 second before and 0.25 seconds after a trigger were considered 'hits'. The most accurate detections were 96% hits and 7% false positives (subject DV); 89% hits and 0% false positives (subject TS, session 2), and 92% hits and 11% false positives (subject TS, session 1). Several possibilities for improved detection accuracy were examined. Templates containing only the peak of the average ERP produced significantly reduced detection accuracy. Detection with variable length templates that included all of the components of the average ERP did not differ significantly from that with fixed duration templates. ECoGs filtered with lowpass cutoff frequencies at 2.5 and 5 Hz produced significantly improved detection accuracy compared to unfiltered ECoGs and ECoGs filtered at higher cutoff frequencies. The relative ease with which ERPs were identified and the high accuracy obtained for some ECoG channels indicates that a direct brain interface based on the use of cross-correlation to detect ERPs related to specific voluntary actions is feasible and worthy of further development.Ph.D.Applied SciencesBiological SciencesBiomedical engineeringNeurosciencesUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/130496/2/9732101.pd