Journal of Neuroscience Methods 147 (2005) 15–21 High resolution parametric description of slow wave sleep Piotr J. Durka a, ∗, 1 , Urszula Malinowska a, 1 , Waldemar Szelenberger b , Andrzej Wakarow b , Katarzyna J. Blinowska a, 1 a Warsaw University, Institute of Experimental Physics, Department of Biomedical Physics, ul. Ho˙ za 69, 00-681 Warszawa, Poland b Medical University of Warsaw, Department of Psychiatry, ul. Nowowiejska 27, 00-665 Warszawa, Poland Received 12 November 2004; received in revised form 7 February 2005; accepted 23 February 2005 Abstract We propose a new framework for quantitative analysis of sleep EEG, compatible with the traditional analysis, based upon adaptive time– frequency approximation of signals. Using a high resolution description of EEG rhythms and transients in terms of their time occurrence and width, frequency and amplitude, we present a detailed detection and parameterization of delta waves, including also the time occupied by each delta wave—a parameter inaccessible directly by previously applied signal processing methods. To validate the proposed parameterization, we construct a simple detector of sleep stages 3 and 4, based explicitly upon the classical criteria related to delta waves. To properly compare its performance to the inter-expert agreements and other expert systems, we sort out and discuss the methodology of reporting concordance in this context. Since the proposed parameterization proves to be compatible with the visual analysis of EEG, we can derive new variables for quantitative analysis of EEG patterns recognized for decades. As examples, we present a continuous description of delta waves and sleep spindles in the overnight sleep, and compare results to the traditional FFT-based estimates. © 2004 Elsevier B.V. All rights reserved. Keywords: EEG slow waves; Delta waves; Sleep spindles; Matching pursuit; Time–frequency; Hypnogram; Contingency tables; Reporting concordance 1. Introduction The most widely used interpretation of polysomnographic recordings relies on division into sleep stages conforming to the standard criteria, summarized by Rechtschaffen and Kales (1968) (R&K). These rules were originally intended as a reference method; nevertheless, for over 35 years they have been used as a gold standard. This naturally causes problems, especially for describing sleep of elderly (Larsen et al., 1995), patients with epilepsy (Marzec and Malow, 2003), and sleep disordered breathing (Norman et al., 2000). Also, stages described by R&K do not reflect many relevant features of sleep EEG. Among the most needed extensions, Billiard (2000) quotes: ∗ Corresponding author. Tel.: +48 22 5532126; fax: +48 22 6226154. 1 Partially supported by Committee for Scientific Research (Poland). E-mail addresses: durka@fuw.edu.pl (P.J. Durka); ula@fuw.edu.pl (U. Malinowska); wald@psych.waw.pl (W. Szelenberger); awakar@psych.waw.pl (A. Wakarow); kjbli@fuw.edu.pl (K.J. Blinowska) URL: http://brain.fuw.edu.pl (P.J. Durka). • a finer time scale than the arbitrary division into 20–30 s epochs, • a measure of spindle intensity, • a quantitative determination of EEG components in the low frequency range. We propose a unified solution to these issues by means of a repeatable procedure, based upon explicit parameterization of local EEG patterns. Waveforms present in the signal are not only described in terms of their frequency and amplitude (energy), but, contrary to the previously applied methods, their exact time positions and durations are also determined. This approach conserves compatibility with the visual EEG analysis. To test the automatic parameterization of delta waves, in this paper, we evaluate detection of the deep sleep stages (3 and 4) based directly upon the classical R&K (Rechtschaffen and Kales, 1968) criteria. Results are compared to visual de- tection and scoring, including the inter-expert concordance. Different ways of reporting concordance of detection are summarized and discussed. 0165-0270/$ – see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.jneumeth.2005.02.010