Quantitative EEG Analysis of Sleep Chronophysiology: A Comparison Between Root Mean Square (RMS) Estimation And Fast Fourier Transformation (FFT)

Abstract

Few formal comparisons between fast Fourier transform (FFT) and digital period-amplitude (DPA) analyses of all-night electroencephalographic (EEG) data have been reported. However, the uses of both approaches have been reported widely in the sleep literature. Objectives.We sought to quantitatively model the temporal dynamics of sleep EEG using FFT and DPA derivations of spectral estimates applied to the same set of raw data, then to determine the degree of statistical similarity or difference between the two sets of results. Approach.Sleep EEG data (C3-A2 lead, 272 samples/sec, band-passed at 0.5-70 Hz) from eleven, normal, male subjects (23-41 yr) were acquired from the second of three nights spent in the laboratory. The raw data were subjected to an FFT and combined into delta, theta, alpha, sigma, and beta bands and 30-second epochs. This produced band-specific estimates of signal power. The DPA data reduction was accomplished by finding the half-wave peak voltage of the bandpass filtered signal between zero crossings, calculating its band mean for the 30-second epoch and dividing the band mean by the square root of 2. This gave band-specific root mean square (RMS) estimates of signal amplitude (the square root of power). Epochs containing artifact were identified visually and not used. Three-minute epochs were created from the means of 30-second epochs. The Spearman rank-correlation coefficient was calculated within bands and within subjects for each of the FFT and DPA 3-minute epochs. A 5-level (bands) Friedman nonparametric 1-way analysis of variance was calculated for the correlation coefficients with the null hypothesis of equality of correlation across bands. Results.Chronophysiologic dynamics of the EEG appeared qualitatively quite similar, with some minor exceptions, across the two approaches for all five bands. The average correlations for the five bands were .942, .849, .803, .817, and .866, respectively. The Friedman statistic was highly significant. Post hoc analysis showed the correlation within the delta band to be significantly higher than within the theta, alpha and sigma bands. Temporal reciprocity across the night between delta and beta patterns was quite evident in both the FFT and DPA data. Conclusions.The two different derivations of spectral estimates of a sleep EEG signal, FFT and DPA, provided highly similar qualitative and quantitative results. The finding of an inverse temporal relationship between delta and beta activities with both approaches replicated a similar finding reported by another laboratory. The inter-method reliability of the two approaches supported the usefulness of quantitative analyses of sleep EEG data. j j, /fv? ■ lah , 1 ',;~x