EEG analysis using moving average-type neural network

Abstract

The brain is one of nonlinear systems. EEG (electroencephalogram) is widely used to monitor the brain condition. Frequency analysis method is given as general EEG analysis method. DR analysis is to evaluate the deviation ratio of power spectrum of arbitrary frequency band. On the other hand, neural networks are used to analyze nonlinear systems because of their ability to capture the dynamics of complex system. In this study, we propose the application of moving average-type neural networks (MANN) to analyze the EEG and compare the result of MANN analysis with the one of DR analysis. In this time, we used the randomness to optimize the number of input units of MANNs. In MANN analysis method, the measured EEG is divided into overlapped period 20 seconds. A MANN is trained to predict the EEG using three previous samples in each period. After training, the connecting weights of each MANN are compared using the inner product to evaluate changes in brain condition. We used measured EEG in closing/opening eyes to analyze. According to the results, the usefulness of MANN analysis was confirmed. And MANN analysis has possibility to be able to apply to analysis of the other EEG data or the other nonlinear systems.