TIME AND TIME-FREQUENCY METHODS IN THE ANALYSIS OF HEART RATE VARIABILITY M. G. Tsipouras and D. I. Fotiadis Unit of Medical Technology and Intelligent Information Systems Dept. of Computer Science, University of Ioannina, 45110 Ioannina, Greece 1. SUMMARY Several time and time-frequency methods are applied to the tachograms obtained from arrhythmic ECG recordings. Small segments of the tachograms are used for feature extraction, which are fed into a neural network to classify them as normal or arrhythmic. The methods are tested using the MIT-BIH recordings and results are presented for all combination of features in terms of the obtained sensitivity and specificity. 2. INTRODUCTION Arrhythmias are disorders of the regular rhythmic beating of the heart. The Electrocardiogram (ECG) of healthy individuals in resting conditions exhibits periodic variation in RR intervals, corresponding to respiratory activity, known as Respiratory Sinus Arrhythmia (RSA). Non-natural arrhythmias can take place in a healthy heart and be of minimal consequence, but they may also indicate a serious problem and lead to heart disease, stroke or sudden cardiac death [1]. Heart Rate Variability (HRV) refers to the beat-to-beat alterations of heart rate. HRV believed to be a good marker of the individual’s health condition and heart diseases [2,3]. Therefore HRV analysis became an important tool in cardiology. Time domain analysis (statistical measurements, geometrical evaluation [3,4,6-8]) and frequency domain analysis [3,6-8] are the most commonly used methods. Non-linear – chaotic analysis has also been used [3,5,6-8]. Time domain analysis provides essential but not detailed information for HRV. Time- Frequency (TF) analysis, which is based on TF distributions, is a more detailed analysis, which provides with non-stationary information of the HRV. Several TF distributions have been used for TF analysis. The Wigner Ville Distribution has been used for the identification of severe brain stem injury and postural tachycardia syndrome [9-13]. Keselbrener et al. used Selective Discrete Fourier Transform (SDA) and Short Time Fourier Transform (STFT) for cardiovascular control and fast vagal response [14-16]. Chan et al. used Wigner Ville Distribution (WVD) and Smoothed Pseudo Wigner Ville Distribution (SPWVD) for Cheyne- Stokes oscillation detection [17-19]. Bentley et al. used the Choi-Williams Distribution (CWD) for classification of native and bioprosthetic heart valve sounds [20].