Use of SSA and MCSSA in the Analysis of Cardiac RR Time Series

A new preprocessing procedure in the analysis of cardiac RR interval time series is described. It uses the singular spectrum analysis (SSA) and the Monte Carlo SSA (MCSSA) test. A novel feature of this preprocessing procedure is the ability to identify the noise component present in the series with a given probability and to separate the time series into a trend, signal, and noise. The MCSSA test involves testing whether the modes obtained from SSA can be generated by a noise process leading to separation of the noise modes from the signal. The procedure described here does not discard or modify any sample in the record but merely separates the time series into a trend, signal, and noise, allowing for further analysis of these components. The procedure is not limited to the length of the record and could be applied to nonstationary data. The basis functions used in SSA are data adaptive in that they are not chosen a priori but instead are dependent on the data set used, increasing flexibility to the analysis. The procedure is illustrated using the RR interval time series of a healthy, congestive heart failure, and atrial fibrillation subject. 1. Introduction Singular spectrum analysis (SSA) [1–4] is an analytical tool that is used in time series analysis. Although it has been used widely in the analysis of environmental data [1–6] its use in biomedical signals has not received much attention. Some of biomedical signals that have been used to illustrate SSA are electroencephalogram (EEG) signals collected during evoked potential studies [7], ultrasound [8], and the electrocardiogram signal (ECG) [9, 10]. In the latter it was used to separate the fetal ECG signal from the maternal signal and in the separation of the heart sound artifact from the respiratory signals. This paper is an attempt to illustrate its value in the analysis of the cardiac RR time series data. The main focus in this paper is on preprocessing the cardiac RR interval series. It proposes an alternate procedure to the use of wavelet analysis for trend removal and the use of impulse rejection filter to remove artifacts [11]. Measured cardiac data contains a large amount of noise and nonlinear trend which require separation before any statistical assessment on the signal can be carried out. The method proposed here uses SSA, followed by the Monte Carlo singular spectrum analysis test (MCSSA) [1, 12]. It is a novel technique not only to separate the signal into various SSA modes but also to identify the SSA modes which correspond to noise. The MCSSA test [1, 12] involves testing the SSA