EEG based artificial learning of motor coordination for visually inspired task using neural networks

Damage in parietal and/or motor cortex of the brain can lead to inability in proper visuo-motor coordination, hampering movement planning and execution. The objective of this work is to predict joint coordinates of hand by sequential prediction of the parietal and motor cortex Electroencephalogram (EEG) features from their occipital counterparts using artificial neural networks (ANNs). EEG signals during hand movement execution are acquired from occipital, parietal and motor cortical regions and the joint coordinates of hand are acquired using Kinect sensor. The acquired EEG signals are preprocessed followed by extraction of wavelet features and selection of the best features using Principal Component Analysis. The EEG features originating from one brain region are mapped to the features of another brain region using regression analysis on artificial neural networks with Back Propagation learning. The mapped motor cortical EEG signals are finally used to predict the hand joint coordinates using Back Propagation learning based ANN. The performances of various weight adaptation techniques for Back Propagation learning are evaluated. Regression analysis results indicate that Levenberg-Marquardt optimization based weight adaptation performed best in terms of mean squared error, slope of the best linear fit and correlation coefficient between the original values and predicted results.