An HMM-PCA approach for EEG-based brain computer interfaces (BCIs)

Official URL: http://192.168.1.20/record=b1301368 (Table of Contents)

Electroencephalography (EEG) based Brain-Computer Interface (BCI) systems are a new development in the field of applied neurophysiology. This new approach has been made possible thanks to progress in EEG analysis and in information technology which has led to a better understanding of psychophysical aspects of the EEG signals. BCI systems enable information flow from the brain directly to the outside world. For widespread use of brain signals for such objectives, effective signal analysis and pattern recognition techniques are needed. In this thesis, we have developed a new technique based on hidden Markov models, and have demonstrated the effectiveness of our algorithms both on a standard dataset and on the data that we have collected in our laboratory. We have used HMMs with AR features combined with PCA to classify two and four class single trial EEG data recorded during imagination of motor actions type of BCI experiments. Results were compared with previous HMM based BCI classifiers and Mahalanobis distance classifier fed with two different state-of-theart EEG features.