The paper investigates the performance of multilayer back propagation neural network (MLP-BP NN)) with resilient training method for discrimination of five mental tasks. The principal component analysis (PCA) was used for feature extraction of the relevant frequency bands from raw electroencephalogram (EEG) signals. The five mental tasks used were relaxed, movement imagery, geometrical figure rotation and arithmetic task (trivial and nontrivial multiplication).

Mental task classification by recognizing electroencephalograph (EEG) patterns is an important and challenging biomedical signal processing problem. Such a classification can be utilized effectively by severely handicapped persons to communicate with their external surroundings [1-7]. It is well documented that proper feature extraction and classification methods are the key to deciding the accuracy and speed of brain computer interface (BCI) systems. Artificial neural network (ANN) has been more widely accepted as the best classification method to distinguish various mental states from relevant EEG signals [8-15]. In this study, principal component analysis (PCA) method was used to capture the information out of five different mental tasks. The coefficients of PCA were used as the best fitting input vector for MLP-BP NN. MLP- BP NN with resilient training method was used to compare the performance in discriminating the five mental tasks. Eight channel EEG systems from ten healthy subjects were used for the present study.

Ten healthy right-handed subjects of (Age: mean 23 Y, Gender: 9M and 1F) having no sign of any motor neuron diseases were selected for the study. A proforma was filled in with the details of name, age and education level. The participants were selected for their availability and interest in the study. EEG data was collected after taking a written consent on willingness for participation. Full explanation of the experiment was provided to each of the participants.

EEG data used in this study was recorded on a Grass Telefactor EEG Twin 3 Machine available at the Department of Neurology, Sir Ganga Ram Hospital, New Delhi. The EEG recording for the ten selected subjects was done for five mental tasks for five days. The data was recorded for 10 seconds during each task and each task was repeated five times per session per day. Bipolar and referential EEG was recorded using eight standard positions C₃, C₄, P₃, P₄, O₁ O₂, F₃ and F₄ by placing gold electrodes on scalp, as per the international 10-20 standard system of electrode placement as shown in Figure 1. The settings used for data collection were: low pass filter 1Hz, high pass filter 35 Hz, sensitivity 150 microvolts/mm and the sampling frequency was fixed at 400 Hz. The reference electrodes were placed on the ear lobes and ground electrode on the forehead. The EOG (Electrooculargram) being a noise artifact, was derived from two electrodes placed on the outer canthus of the left and right eye in order to detect and eliminate eye movement artifacts.

Science and Technology Journal, Electroencephalogram (EEG), Principal Component Analysis (PCA, Neural Network, Artificial Neural Network (ANN, Electrooculargram, Fast Fourier Transform , FFT, Psychophysiology, Biomedical Computing, Neural Engineering, EOG , Electrooculargram.