279 SCRIPTA MEDICA (BRNO) – 79 (5–6): 279–288, December 2006 matching pursuit decomposition for detection of frequency changes in experimental data – application to heart signal recording analysis BARDOňOvá J. 1 , PROvAzNík I. 1 , NOvákOvá M. 2 1 Department of Biomedical Engineering, Brno University of Technology, Brno 2 Department of Physiology, Faculty of Medicine, Masaryk University, Brno Received after revision November 2006 Abstract The time-frequency analysis is a powerful tool to describe the nature of non-stationary biological signals. Short Time Fourier Transform and Wavelet Transform are well-known analysis tools used in the experimental field. This paper deals with another method, Matching Pursuit (MP) decomposition. This method decomposes a signal into an optimal linear expansion of waveforms, which are functions previously defined in a dictionary, and thus extends capability of traditional tools. The MP method is applied to study changes of energy of ECG signal frequency components dur- ing three experimental phases. The experiments include application of voltage-sensitive dye (vSD) needed for non-invasive optical mapping from heart surface. Our previous Langendorff perfused heart experiments suggested shape changes in ECG signals in time-frequency domain caused by application of vSD di-4-ANEPPS. In this study, the heart cycles were decomposed by MP in all phases (control, loading and washout period). The histograms of the relative frequency of waveforms resulting from MP were computed to show frequency details for each experimental phase. The study shows significant shifts of majority energy frequency components during loading and their recovery after washout. Further, MP confirmed subtle frequency changes within QSR complexes during the experiment. key words voltage-sensitive dye di-4-ANEPPS, ECG signal, Wavelet analysis, Matching Pursuit decomposition INTRODUCTION various heart diseases can be studied from the recordings of a range of signals reflecting heart function or anatomy. Action potentials, electrocardiograms, ultra- sound images, intracardial pressure are examples of the most often used recordings. Their analysis significantly contributes to clinical diagnostics and to basic cardio- logy research. Today’s methods of biological signal analysis involve a number of sophisticated and complex mathematical approaches allowed by the use of high-per- formance computers. However, the choice of the appropriate method is difficult as the electrophysiological phenomena to be detected are usually expressed by subtle changes in the recordings.