Computers in Biology and Medicine 37 (2007) 1502 – 1510 www.intl.elsevierhealth.com/journals/cobm Combining classical HRV indices with wavelet entropy measures improves to performance in diagnosing congestive heart failure Yalçın ˙ I¸ sler ∗ , Mehmet Kuntalp Electrical and Electronics Engineering Department, Dokuz Eylül University, ˙ Izmir 35160, Turkey Received 21 July 2006; accepted 23 January 2007 Abstract In this study, best combination of short-term heart rate variability (HRV) measures are sought for to distinguish 29 patients with congestive heart failure (CHF) from 54 healthy subjects in the control group. In the analysis performed, in addition to the standard HRV measures, wavelet entropy measures are also used. A genetic algorithm is used to select the best ones from among all possible combinations of these measures. A k-nearest neighbor classifier is used to evaluate the performance of the feature combinations in classifying these two groups. The results imply that two combinations of all HRV measures, both of which include wavelet entropy measures, have the highest discrimination power in terms of sensitivity and specificity values.  2007 Elsevier Ltd. All rights reserved. Keywords: Heart rate variability; Genetic algorithm; k-Nearest neighbor rule; Feature selection; Congestive heart failure; Wavelet entropy 1. Introduction The analysis of heart rate variability (HRV) is a standard method for studying the role of autonomic nervous system (ANS) in heart control. By the analysis of the oscillations be- tween consecutive heart beats, various kinds of defects can be detected [1]. Although HRV has been the subject of numer- ous clinical studies investigating a wide spectrum of cardio- logical and non-cardiological diseases and clinical conditions, a general consensus on the practical use of HRV has been reached only in two clinical scenarios: depressed HRV has been used as: (i) a predictor of risk after acute myocardial infarc- tion (AMI) and (ii) an early warning sign of diabetic neuropa- thy. In other clinical conditions and diseases, depressed HRV has also been observed in patients suffering from dilated car- diomyopathy (DCM), congestive heart failure (CHF), fetal dis- tress conditions, and obstructive sleep apnea [1–11]. Majority of these studies use HRV measures as predictors of the risk of ∗ Corresponding author. Tel.: +90 232 4127180; fax: +90 232 4534279. E-mail addresses: islerya@yahoo.com (Y. ˙ I¸ sler), mehmet.kuntalp@deu.edu.tr (M. Kuntalp). 0010-4825/$ - see front matter  2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.compbiomed.2007.01.012 mortality (prognosis) [12]. Only a few studies are focused on using HRV measures for diagnosis purpose [9,13]. The HRV data is obtained from electrocardiogram (ECG) records in which the heartbeats are recognized through QRS complexes. Since R peaks in QRS complexes indicate the ven- tricular contractions, the beat instants are taken at these points and consequently the beat-to-beat intervals are determined as the length in time from one R wave to the next one. Therefore, the term “beat-to-beat interval” refers to RR interval which is an unevenly sampled data. The RR intervals are sometimes re- ferred to as NN (normal-to-normal) intervals when they are resulting from normal sinus rhythm [1]. In 1996, the Task Force of the ESC/NASPE published stan- dards in HRV analysis proposing several time and frequency parameters based on short-term (5-min) and long-term (24- h) HRV data [1]. Time-domain measures are simple statisti- cal methods to calculate from both long- and short-term raw HRV data. The time-domain measures used in HRV analysis are given in Table 1. In addition to these statistical measures, HRV analysis is also done using FFT-based power spectral density (PSD) mea- sures. This type of analysis provides the basic information on how power distributes as a function of frequency. Three main