Specificity and reliability of measures of the EEG power spectrum

Abstract

Recent studies indicate that the genetic determinants of individual differences in cortical structure and function may be reflected in the electroencephalogram. To investigate a possible genetic influence on the EEG it is necessary to derive a measure of the EEG that is specific, unique and reliable. This thesis investigates a number of measures that are a candidates for a genetic investigation of the EEG.Firstly, an investigation was made into the point-D2i and correlation dimension, both measures purporting to index the complexity of the cortical organisation giving rise to the EEG. The major focus was on the point-D2i, but both it and the correlation dimension were shown to be unsuited to studying EEG genetics.Secondly, statistical properties of a number of indices derived from the log power spectrum of the EEG were described. The EEG power spectrum is most commonly analysed in terms of the classical delta, theta, alpha and beta bands. In this thesis analysis was concentrated on the individual frequency bins of the spectrum, with a limited study of the classical bands for comparative purposes. Comparison was also made between a number of different methods for averaging spectra: the classical method of averaging spectra aligned to the first bin of the power spectrum, as well as an averaging methodwhere spectra were aligned to various estimators of the dominant frequency of the alpha band prior to averaging.It was shown that the peak Alpha frequency and the gravity frequency were normally distributed in the average, and both showed high reliability within a single test session and across two recording sessions spaced about 3 months apart. Topographic variation in reliability or frequency was not marked for the gravity frequency. The peak Alpha frequency varied topographically, being highest occipitally and declining frontally. Frontal reliability was also lower than for the central, parietal and occipital regions.The skew and kurtosis of individual bins of the power spectrum showed little topographic or frequency dependant variation, except for lower values in and around the alpha band. However, the reliability of individual frequency bins of the log power spectrum showed both topographic and frequency dependence, as did the stability of the local shape of the power spectrum.The major conclusion of the thesis is that centering the averaging of individual spectra to the peak alpha frequency provides the best method for isolating the process generating power at the peak alpha frequency. The high reliability and temporal stability of log power at the peak alpha frequency when using the alignment method indicates that this measure is a promising candidate for investigating the genetic basis of the EEG.