Brain Computer Interface Design Using Mental Tasks Classification

In this paper, a bi-state Brain Computer Interface (BCI) using neural network (NN) classification of electroencephalogram (EEG) signals extracted during mental tasks has been designed. The output of the BCI design could be used with a translation scheme such as Morse Code for paralysed individuals to communicate with their external surroundings. In the experimental study, EEG signals from 5 mental tasks were recorded from 4 subjects and combinations of 2 different mental tasks were studied for each subject. Three different feature extraction methods were employed in the BCI design: 6 th order autoregressive (AR) coefficient computed with Burg’s algorithm, power and asymmetry ratio from delta, theta, alpha, beta spectral bands, and power spectral density (PSD) values. The NN utilised for classifying the features were Multilayer Perceptron architecture with Backpropagation training (MLP-BP) and Simplified Fuzzy ARTMAP (SFAM). Comparisons of classification performances were made among the 3 different feature extraction methods and the NNs. The results indicated that 6 th order AR coefficients gave the best performance, in addition to requiring the least amount of time to train and test. In general, MLP-BP performed better than SFAM. The results also showed the importance of selecting suitable mental task combinations to maximise the BCI output for each subject because of the varying NN classification performances for different mental tasks.