Structural analysis of human electroencephalograms : an exploratory study

A structural analysis technique for human electroencephalograms (EEGs) was investigated. The analysis is based on the assumption that an EEG can be considered as being composed of a finite number of short elementary patterns. Subtle changes in an EEG may be detected by studying the probabilities that an elementary pattern of type i is followed by one of type j, that is, constructing a transition probability matrix may be an appropriate way to study EEG structures. A number of experiments were performed with artificially generated data to study the behavior of the techniques for the estimation and comparison of the transition matrices. It was found that the data length required to reliably estimate the matrices depends on the number of states and zero entries of the matrices. For estimating a matrix without zero entries, a data length of about six to eight times the number of entries is required. The changes of the transition probabilities that may occur as a result of abnormalities or normal fluctuations in the psychophysiological condition were studied using visually scored sleep-stage sequences and actual EEG tracings consisting of normal and pre-seizure EEGs. For the sleep-stage sequences, it was found that the structural analysis agrees with the sleep somnograms and reveals the changes between different sleep patterns. For the EEG tracings, the structural analysis was used to describe the normal EEG variability and to detect any changes during pre-seizure periods. The results were compared to a parameter averaging approach, and to a visual interpretation of the EEGs. It has been shown that the structural analysis reveals visually confirmable EEG changes, although these changes may not necessarily be related to an impending seizure. The preliminary results presented in this dissertation indicate that the structural analysis has promises for the future, but many aspects need further study before it can become a clinically useful tool.