|
Institute of Behavioral Sciences, Semmelweis University, Budapest, Hungary
The power spectra of the NREM phases of sleep EEG is characterized by a high individual variability. This individual EEG-trait is substantially invariant across several consecutive nights characterized by large experimentally induced changes in sleep architecture. This feature is considered as a “fingerprint”. The common way of converting this individual shape to numbers is the integration of the curve in some frequency ranges. The frequency ranges are chosen arbitrarily by the average mean frequencies of the EEG activities. As the individual variations in the mean frequencies are often outside of these ranges, scientists and clinicians loose potentially relevant information when using spectral analysis of the sleep EEG. We present a new method of parametrization of the shape of the average power spectrum by curve fitting based on the 1/f approximation of the EEG. According to the theory of Freeman the average power spectrum is divided into two parts: a fitted linear in the log-log scale as expressing stochastic activity by the y-crossing and the power of the 1/f function and the sum of a defined number of Gaussians with different middle frequency, width and height. Based on our analyses on a database of 45 healthy subjects’ second night laboratory sleep the fitted parameters provide a reliable characterization of the individual shape of the periodogram, reflecting at the same time the common spectral features (coloured noise+peaks) with superior analytic precision, and correlating with the outputs of other procedures, like the individual adjustment method (IAM) of sleep spindle analysis as well as with the age of the subjects. The method provides a parametrization of the sleep EEG fingerprint, as well as a potential start-up for the expansion of the concept of endophenotypes in psychiatry and sleep medicine. |