Improving classification of EEG signals for a four-state brain machine interface

Neural network classifiers are one among the popular modes in the design of classifiers for electroencephalograph based brain machine interfaces. This study presents algorithms to improve the classification performance of motor imagery for a four state brain machine interface. Dynamic neural network models with band power and Parseval energy density features are proposed to improve the classification of task signals. Motor imagery signals recorded noninvasively at the sensorimotor cortex region using two bipolar electrodes are used in the study. The performances of the proposed algorithms are compared with a static neural classifier. Average classification performance of 97.7% was achievable. Experiment results show that the distributed time delay neural network model out performs the layered recurrent and feed forward neural classifiers.