The use of advanced information processing methods in EEG analysis
Stephen Roberts, Iead A Rezek, W.D. Penny, Richard Everson · 1998
this paper the problems of noise and artifact removal, nor the problems of modelling the passage of the EEG from cortex to scalp, but instead will concentrate on the EEG as is and suggest that, even if such problems are not solved, the EEG contains significant objective information regarding cortical functioning. We focus upon two main areas of information processing, namely information extraction from an EEG record (feature extraction and representation) and pattern recognition and inference. Why `advanced' methods? Traditional EEG analysis is based upon signal power in a series of frequency bands, 0.5-4Hz the delta band, 8-12Hz the alpha band etc. Whilst such a paradigm is well-suited to analysis by eye it is not necessarily the `best' way to perform an automated analysis. In the case of multiple channels of information, for example, it is not immediately clear how information about the spectral content of one channel relates to that in other channels. Over the last decade advances in computing have enabled more sophisticated methodologies to tackle the large data processing tasks involved in EEG analysis. We focus on two techniques; firstly feature encoding using measures of stochastic complexity and secondly the use of flexible models (in particular `neural' networks) for pattern recognition. Stochastic complexity Whilst frequency-based changes clearly do occur in EEG recordings, we have argued that they are manifestations of changes in synchronisation in populations of neurons. We observe, for example, that determination of brain state is complicated by the fact that the effect on the EEG's frequency spectrum by changes in state differs from individual to individual. Frequency-based methods, therefore, cannot be guaranteed to be robust. We argue that methods which ...