spectrotemporal Mapping of signal averaged ecg in
taiwanese chronic renal failure patients

Chun-Cheng Lin¹, Ten-Fang Yang^2,3, Chih-Ming Chen¹, Ing-Fang Yang^3
1 Department of Electrical Engineering, National Taiwan University of Science and Technology,
² Graduate Institute of Medical Information, Taipei Medical University,
³ Jen-Chi General Hospital, Taipei, Taiwan.

Abstract: The purpose of the present study is to develop the spectrotemporal mapping (STM) for signal averaged electrocardiograms (SAECGs) of normal Taiwanese and CRF patients before and after hemodialysis (HD) therapy in order to evaluate the ventricular arrhythmias of chronic renal failure (CRF) patients, and the effects of HD therapy on STM analysis of SAECGs.

The quantitative analysis of spectrotemporal domain SAECG was performed based on a cross-correlation function to define a normality factor (NF) in evaluating the variations on high frequency components of SAECG. The CRF patients before HD therapy have a higher NF ( %) with significant statistical difference (p < 0.05). After HD therapy, the NF of CRF patients has reduced to  % without significant statistical difference as compared to normal (p > 0.05). Hence, the application of SAECG STM technique in the current study would effectively separate normals and CRF patients, and reflect the effects of HD. This might be of clinical significance in the evaluation of cardiovascular status of CRF patients.

Introduction

The detection of ventricular late potentials (VLPs) using SAECG has been an important and noninvasive technique to evaluate patients with high risk ventricular arrhythmia [1]. In recent years, there have been several researches using SAECG for the evaluation of high risk ventricular arrhythmias in CRF patients under HD [2-3].

Vector magnitude analysis developed by Simson at 1981 has been widely used in time domain analysis of SAECG. However, several reports revealed that the residual noises would still affect the final analysis results; even the noise level has been reduced below the standard value [4-6].

In 1989, Haberl et al [7] introduced STM, which analyses the frequency content of VLPs based on moving window techniques and is independent of the noise interference. Although the STM has been adopted by several research groups in the analysis of frequency domain SAECG, the standard analysis procedure has not yet been established [6].

The purpose of the present study is to develop the STM for normal Taiwanese and CRF patients before and after HD therapy in order to evaluate the ventricular arrhythmias of CRF patients, and the effects of HD.

MATERIALS AND METHODS

Materials

There were 52 normal Taiwanese (N) (51 men and 1 woman, aged  years old) recruited for the present study. They were students and teaching staff from Chin Min College of Technology and Commerce. Twenty-six CRF patients (13 men and 13 women, aged  years old) undergoing maintenance HD at Jen-Chi General Hospital were also included in this study. CRF patients and normals with evidence of bundle branch block or intraventricular conduction delay were all excluded from the present study.

Bipolar recording system

The high resolution ECGs were recorded using a commercially available Simens-Elema Megacart^® machine. The high-resolution measurements of CRF patiens were separately performed 30 minutes before and after HD. A bipolar, orthogonal X, Y and Z lead system was used [5]. A sample of 10 minutes raw ECG data with 12-bit resolution at 2 kHz was stored on computer hard disk for subsequent analysis.

Analysis of High-Resolution ECG

According to the 1991 ESC, AHA and ACC Task Force [5], SAECGs were performed at the root mean square value (RMS) noise level set below 0.7 microV by a 40-250 Hz bi-directional, high pass Butterworth filter. In time domain analysis, onset and offset were determined from the vector magnitude (VM).

The moving windows were applied in STM to observe the variations of high frequency components of late QRS complex. In this study, the offset was obtained from VM analysis as a reference point. The Blackman Harris window with 80 ms length was adopted. The first segment started 28 ms after the end of QRS complex; the subsequent segments were shifted by increments of 2 ms retrogradely into the QRS complex. The last segment therefore began 20 ms before the QRS offset. After windowing, each segment passed through a high pass Butterworth filter with 10Hz cutoff frequency for reducing the interference effect on high frequency component from the higher amplitude low frequency component.

For each windowed and high pass filtered segments, the spectral estimation based on fast Fourier Transform (FFT) was performed by the following definition of periodogram spectral estimator to calculate the power spectral density (PSD):

where N is the segment length, 0 N-1. PSD was represented as dB value by . When dB value is 0, the PSD only equals 1. The PSD < 0 dB was considered as non-significant and set as 0 dB. All estimated spectra were aligned and displayed as a 3D STM.

In order to quantify the variations of high frequency components in STM, a reference spectrum was defined as the average spectrum of segment 1 to 5 and the spectrum of segments 1-25 were compare with this reference spectrum by corss correlation in the frequency range  to 150 Hz ( is set as the frequency when the high frequency component of reference spectrum is 10 dB,  is about 40). According to these spectral correlations, a normality factor (NF) was defined: the mean of the correlation coefficients of segments 21-25 multiplied by 100. Figure 1 (a) and (b) are the STM of a CRF patient before and after HD therapy in lead X. That means the greater variation of high frequency component of the STM, the lower the normality factor.

Statistical methods

All statistical analysis was done with the commercial Microsoft Excel® software. Data were presented as mean standard deviation (SD). Student’s t test was used for comparing means of two independent variables and F test was used for the variance comparisons between parameters. Statistical significance was defined as p values < 0.05.

Results

The NF results obtained from STM for normals, CRF patients before and after HD are shown in table 1. Comparing the NF results of Y and Z lead SAECGs and the mean NFs of three leads, the CRF patients before HD have higher NFs than normals, and the NF reduced to the range of normals after HD. These differences are statistically significant.

DISCUSSION

The fixed frequency range of 40 to 150 Hz used in calculating correlation coefficients of STM of SAECG by Haberl et al [7] was modified by using to150 Hz in the present study. Two main reasons for applying this technique are: (1) The dB values originated from 40 to 150 Hz in certain reference spectra are at the value zero, the correlation coefficients can not be calculated, (2) when there is higher dB value at 40Hz spectrum, higher correlation coefficients would be obtained even with greater variations in high frequency components. Therefore, the effect of high frequency variations cannot be accurately evaluated.

In conclusion, the differences of frequency domain SAECG parameters between CRF and normals were well illustrated by adopting the newly locally developed spectrotemporal analytic techniques. The greater variations of NF in frequency domain SAECG before HD and normals were significantly reduced after the HD. This result was observed from the present study. There were also lower high frequency component variations of late QRS complex in CRF patients before HD. This phenomenon cannot be interpreted with the clinical electrophysiological changes and needs further investigation.

(a)

(b)

Figure 1. STM of a CRF patient in lead X: (a) before HD with 91 % NF, (b) after HD with 71 % NF.

TABLE 1. Normality factors (%) analyzed on STM for normals,
CRF patients before (PreHD) and after HD therapy (PostHD).

	Normals	PreHD	p value
		PostHD
X lead	85 14	87 12	NS
		83 14	NS
Y lead	75 13	79 9	< 0.04
		72 18	NS
Z lead	75 21	83 12	< 0.02
		73 17	NS
Mean of XYZ leads	78 8	83 6	< 0.004
		76 10	NS

References

[1]     T.F. Yang, P.W. Macfarlane, “New sex dependent normal limits of the signal averaged electrocardiogram,” Br Heart J, vol. 72, pp. 197-200, 1994.

[2]     C.C. Lin, T.F. Yang, C.M. Chen, et al., “A novel approach toward Better Signal Averaged ECG frequency Domain Analysis,” Proceedings of Medical Informatics Symposium in Taiwan, 2001, pp. 245-248.

[3]     C.C. Lin, T.F. Yang, C.M. Chen, et al., “The Effect of Noise Levels on the Time Domain High Resolution ECG Analysis,” Proceedings of Medical Informatics Symposium in Taiwan, 2001, pp. 249-252.

[4]     P. Lander, E.J. Berbari, R. Lazzara, “Optimal filtering and quality control of the signal-averaged ECG. High-fidelity 1-minute recordings,” Circulation, vol. 91, pp. 1495-1505, 1995.

[5]     G. Breithardt, M.E. Cain, N. El-Sherif, et al., “Standards for analysis of ventricular late potentials using high-resolution or signal-averaged electrocardiography,” J Am Coll Cardiol,    vol. 17, pp. 999-1006, 1991.

[6]     M.E. Cain, J.L. Anderson, M.F. Arnsdorf, “Signal-Averaged lectrocardiography,” J Am Coll Cardiol, vol. 27, pp. 238-249, 1996.

[7]     R. Haberl, G. Jilge, R. Pulter, et al., “Spectral mapping of the electrocardiogram with Fourier transform for identification of patients with sustained ventricular tachycardia and coronary artery disease,” Eur Heart J, vol. 10, pp. 316-322, 1989.