The widespread availability of powerful personal computers gives a new meaning to the term applied signal processing. Advanced methods are becoming available to users with little mathematical background. Among these methods, adaptive time-frequency approximations of signals are special in more than one respect:
For more than 70 years, EEG have been the most direct trace of thought that we can measure. Recently, electroencephalography seems to have lost importance in favor of new brain imaging techniques---in spite of their only indirect relation to the neuronal signaling, low time resolution, and high cost. Why? Magnetic resonance imaging and positron emission tomography offer results in terms of easily interpretable, computed images; no clinician or neurophysiologist would use the raw signals recorded by the sensors. Funny as it may sound, visual analysis of raw EEG recordings is still the state of the art in clinical electroencephalography--basically unchanged in 70 years.
This book gives blueprints to bridge the gap between the tradition of visual EEG analysis and advanced signal processing. From the same basic ideas, we also derive complete frameworks that open new possibilities in several research paradigms: classical and continuous descriptions of sleep recordings (polysomnograms), microstructure of event-related EEG desynchronization and synchronization, detection/description of epileptic spikes and seizures, pharmaco EEG, and source localization (preprocessing for EEG inverse solutions). The sum of these applications suggests that presented paradigms can unify at least some elements of the art of visual EEG interpretation with advanced signal processing. They also unify advantages of different signal processing methods applied previously in this field. Such common methodological framework may significantly improve the reliability of both clinical and research applications of EEG.

P.J. Durka    Matching Pursuit and Unification in EEG Analysis
Foreword by Fernando Lopes Da Silva Publisher's page Introduction, TOC and sample chapter in PDF

Contents of This Book  (online table of contents)

Digital revolution opens amazing possibilities, but computers do not think for us. To be responsible for the results, we must understand what we are doing. In biomedical sciences, "we" cannot relate only to mathematicians and engineers. Therefore, the first part of this book gives a minimal necessary  background in signal processing, using only plain English and no equations. Starting from basic notions like sampling of analog signals, inner product, orthogonality, and uncertainty principle, through spectral and time-frequency methods of signal analysis (spectrogram and wavelets), we arrive at the idea of adaptive  approximations and the basics of the matching pursuit algorithm. Chapters 6 and 7 summarize major advantages and caveats related to its applications, with references to examples from Part II.
Each of the applications presented in Part II explores some particular and unique feature of the matching pursuit. Starting from the explicit parameterization of signal structures in terms of their amplitudes, time widths and time and frequency centers, through high-resolution and robust estimates of time-frequency energy density and their averages in event-related paradigms, to selective estimates of the energy of relevant structures, which improve the sensitivity of pharmaco-EEG and stability of EEG inverse solutions. Similar to Part I, these presentations are basically equation-free. Software used in these studies is freely available from
For the mathematically oriented readers, Part III introduces formally adaptive approximations and related technical issues, including the mathematical tricks necessary in efficient implementations of the matching pursuit algorithm.