Most current brain computer interface (BCI) methods utilize feature extraction techniques based on some form of signal modeling applied to a single time series of data to identify the state of the EEG system. However, an alternative system identification process is possible, using a temporally specific external stimulus, by building a mathematical model based on observed input and output time series. Transcranial Magnetic Stimulation (TMS) is a more recent field of EEG research that provides one such stimulus. In this paper, we present a new process for identifying the EEG state wherein system identification theory is implemented to model the transformation of the EEG due to a time specific TMS pulse. An AutoRegressive Moving Average with eXogenous input (ARMAX) structure was classified using a Support Vector Machine (SVM) algorithm. The maximum classification accuracy of 88% for a single subject, used a quadratic kernel and alpha frequency, but we also report results from different implementations. The information transfer rate, however, is only 5.1bits/min. This study is the first known to use system identification, and in particular the system identification of the brain's response to a TMS pulse as an index of intention. It provides proof of concept as well as an initial implementation and evaluation of this form of BCI.
展开▼
