14th International Congress
THE "NEW FRONTIERS"
OF ARRHYTHMIAS 2000

Jan. 29 - Feb. 5, 2000
Marilleva, Trento, Italy

RT-212

New algorithm for T wave alternans analysis in Holter recordings

Xavier Vinolas, Salvador Olmos**, Jose L. Alonso, Pere Caminal*, Pablo Laguna**, AntonioBayés de Luna.
Cardiology Department, Hospital Sant Pau, Barcelona, *Centre de Recerca Bioengenieria, Universitat Barcelona, **Facultad Ingenieria, Universidad Zaragoza, Spain

Introduction

TWA detection methods

All detection methods need the following common steps:

1) preprocessing: used to eliminate the non useful components of the ECG signal, for example the variations of the baseline signal)¹⁰;

2) QRS detection and segmentation of ST-ST segment: the step is crucial, because the beginning of the QRS should be similar for all beats¹¹;

3) noise detection: in order to eliminate the noisy beats;

4) low pass filter: used to eliminate the frequency components >40 Hz.

Fast Fourier transform methods

This was the first method used to detect alternans and is the most widely used. But some limitations are present, and some authors prefer the correlation method (see later)¹². The most important factors influencing the FFT result:

– ectopic beats;

– incorrect alignment of QRS complexes;

– noisy beats.

The most important advantages of the FFT method is the simplicity to implement the analysis and the low computer calculations needed. The limitations are those of the FFT analysis: 1) the assumption of a stationary signal, when TWA is clearly non-stationary; 2) the relationship between temporal resolution and frequency resolution. When time interval is reduced (typically to n=128) we can assume that the signal is stationary; but signal resolution is worse with short time intervals; 3) we cannot find TWA amplitude (in microvolts) for each beat; 4) we do not have information regarding the morphology of T wave.

Complex demodulation method

The system uses a high pass filter of the ST-T segments correctly aligned. Two kinds of filters are used:

– deterministic filter

– statistic filter¹⁶.

The most important advantages of this system are: 1) better temporal/frequency resolution compared to FFT analysis and 2) it allows to analyse the morphology of the alternating signal.

Correlation method

This method was proposed by Burattini et al. Summarizing, a template of ST segment (median of the signal in the 128 consecutive signals) is create in the 128 beats sequence. The correlations of the beats with the template are analysed. One of the advantages is that the system is strongest for misalignment of QRS complexes and concentrates all information regarding the ST-T segment in a single number. The use of the median QRS complex as the template could represent a limitation.

KLT based method

The method based on the Karhunen-Loeve transform allows to represent the information regarding signal analysed in a short number of coefficients^20,21. Our group has used this type of analysis for ST changes detection during PTCA^22-26 and alternans detection^27,28. Once the different coefficients representing each beat are obtained, we only represent the component with a period each two beats. We propose the high pass filter statistic method of Capon¹⁶. The method is more solid than the correlation method against noisy beats and allows a better representation of the “morphology”. We can also obtain using this method the duration and amplitude of the “alternating” period.