Analysis of THE normal and THE fractionated conduction DURING ventricular activation USING time-frequency magnitude squared-coherence mapping

P. R. B. Barbosa^1,2, L.C. Bernardino Jr², J. Barbosa-Filho², I. Cordovil^2,3, R.L.M. Sá^2,3, J. Nadal¹
¹Biomedical Engineering Program – COPPE, Federal University of Rio de Janeiro
P.O. Box 68.510, Rio de Janeiro, RJ, 21945-970, BRAZIL. E-mail : ecgar@ecgar.org
²Gama Filho University, Rio de Janeiro, Brazil
³National Institute of Cardiology Laranjeiras

Abstract: Fractionated activity (FEA) in damaged myocardium represents unstable electrical conduction, probably due to dispersion in propagation time-constants. A one-dimensional linear model of continuous velocity waveform transmission is proposed, and a time-frequency magnitude squared-coherence (MSC) mapping of the ventricular activation (VA) of patients with ventricular tachycardia (VT) and normal volunteers was studied. Normal subjects showed high MSC estimates, mainly at start and terminal VA. Reduction of MSC estimate throughout VA was observed in VT, which reflects variation in the content of successive spectral estimates. This finding is indicative of fluctuation in propagation velocity of the ventricular activation due to damaged myocardium.

INTRODUCTION

Prolonged and fractionated electrical activity in damaged myocardium represents a harbinger of the risk of development of life threatening arrhythmia with important prognostic implication. Detected by techniques of averaging and amplification on electrocardiogram (ECG), ventricular late potentials (VLP) are the representative of myocardial fractionation on the body surface and employed for clinical stratification of sustained arrhythmia. Notwithstanding, both in time and frequency domains, current techniques employed for VLP analysis yield low positive predictive value (< 15% in one year)[1]. This figure potentially reduces clinical interest and justifies continuous investigation on the organization of the electrical activity in ventricular myocardium, and the extraction of discriminant parameters for quantification of abnormal electrical activity.

Reentry arrhythmia may occur when the presence of fibrosis intercalated with myocardial fibers induces the reduction of the propagation velocity of electrical stimuli. Consequently, changes will occur in time‑constants and wavefront orientation, with dynamic alteration of the spectral characteristics of the surface ECG. On the other hand, the organization of the ventricular activation (VA) in normal myocardium is supposed to maintain time-constants unchanged for longer periods, yielding smooth propagation.

The objective of this study was to analyze whether the changes of the spectral characteristics of the VA allows stratifying patients (pts) with previous episodes of ventricular tachycardia and normal subjects in sinus rhythm.

METHODS

A linear model of the activation front was proposed to verify if fractionated activity in ventricular myocardium might cause continuous changes in propagation time-constants. The model hypothesized that in an arbitrary period of time during normal VA, the displacement of the propagation waveform in a defined instant is linearly related to the immediately preceding activity (Fig. 1). This would induce spectral harmonics of the ECG signal to be steady, maintaining both amplitude and phase unchanged.

Propagation of the electrical stimulus in normal ventricular myocardium

Figure 1: Linear model of wavefront spreading through the myocardium: an incoming wavefront potential (x(t)) captioned by an exploratory electrode is linearly related to the outgoing one (y(t)) from a given region. X(w) and Y(w) are Fourier transforms and X^*(w) is the conjugate complex. H(w) is the transfer function. In this model, ideal conduction yields local magnitude squared-coherence (MSC(w)) approaching 1.

Population

The study population comprised 30 subjects divided into two groups adjusted by age, gender and anthropometric measurements (Table I): Group 1 (G1)- 15 normal volunteers; and Group 2 (G1)- 14 pts with documented ventricular tachycardia (VT), 8 sustained and 6 non-sustained, and 1 syncope post-myocardial infarction. All individuals were in sinus rhythm, and none had complete bundle branch block. The study protocol was approved by the ethics committee of the National Institute of Cardiology Laranjeiras (INCL) and individuals were admitted after written informed consent.

TABLE I – Demographic, anthropometric and clinical data:

	Group 1	Group 2	p
Age (y.o)^*	49.3±13.4	57.3±13.5	0.13
Gender (F / M)	7 / 8	3 / 12	0.12
Height (m)	1.63±0.09	1.67±0.09	0.40
Weight (kg)	71.7±14.5	67.8±10.2	0.31
APT diameter (m)^**	22.3±2.1	22.6±1.7	0.99
SBP (mmHg)^***	130.3±16.5	132.6±27.3	0.74
DBP (mmHg)^***	82.2±9.0	79.4±14.5	0.49
LVEF (%)^****	71.7±6.8	53.2±13.0	<0.01
Heart rate (bpm)	63.5±11.4	61.5±10.0	0.85
P-wave duration (ms)	101.3±8.3	122.8±23.5	0.01
QRS duration (ms)	91.7±8.8	119.2±17.9	<0.01
RMS40 Lnt (Ln[mV])⁼	1.55±0.21	1.17±0.43	0.03
LAS40 (ms)⁼⁼	28.1±7.1	43.7±19.4	0.03
Positive Late potentials	0	8	-

^*Mean±SD;^**Anterior-posterior thoracic,^***Systolic and Diastolic blood pressure, ^****Left ventricular ejection fraction; ⁼Lnt-Logarithm transform, RMS40– voltage of the terminal QRS; ⁼⁼LAS40– duration of terminal QRS.

ECG Data Acquisition and Processing

The high-resolution ECG signals were acquired using the XYZ Frank leads with 14-bit resolution at 1000 samples/sec, employing a protocol previously described [2]. To reduce baseline noise, beats were aligned and averaged to a final noise <0.6 mV. To assure appropriate synchronization of the VA among all subjects, the averaging beats were aligned so that the point in the ascending limb of the QRS complex corresponding to 0.5 of its amplitude was fixed. This point defined the start of the VA. Employing the histogram of normal RR intervals, beats were select for averaging according to the class corresponding to the shortest RR interval [2].

Intervals of 512 ms of each averaged lead were numerically differentiated to emphasize high frequency content, and analyzed into segments of 64 ms, separated with steps (Dt) of 4 ms. Each segment was further decomposed into 16 ms epochs with 50% overlapping of the preceding epoch. Mean removal and Hanning window tapering of each epoch was performed prior to spectral analysis estimated via FFT algorithm. The auto- and cross-spectral density function estimates of successive segments where used to calculate the Magnitude Squared Coherence (MSC), and respective confidence intervals according to De Sá et al. [3]. For each pair of segments, x(t) and y(t)=x(t+4), MSC function was calculated as shown in Fig. 1, and the mean MSC function estimate (MSC_m) of the three-lead set were mapped in time-frequency domain, during the entire averaged beat.

Echocardiographic, blood pressure and anthropometric measurements were carried out according to INCL routine protocols. VLP was analyzed according to 1996 ACC expert consensus.

The ensemble average MSC_m functions of G1 (MSC1_av) and G2 (MSC2_av) were calculated using Fisher transform, and groups were compared for each time-segment and frequency-segment by the z-test. Additional statistical tests (Table 1) employed Student t and c² tests (Gender). Alfa error was 0.5.

RESULTS

The MSC_m time-frequency map of G2 pts presented a similar pattern to G1 (Fig. 2), except at small low-frequency regions at the onset and offset of the VA which is confirmed by the mapping of the statistical test results (Fig.3).

Figure 2. MCS_m of a subject of Group 1 (left) and Group 2 (right). Note significant reduction in MCS estimates, mainly and the terminal portion of the ventricular activation.

Figure 3. Time-frequency mapping of statistical results of comparison of MCS_av function estimates between Groups 1 and 2 (Left) and the reference ventricular activation (Right). Significant level (p< 0.05) is displayed as light gray. Note reduction of MCS at the boundaries of the ventricular activation (arrows).

DISCUSSION

The analysis of surface ECG using an one-dimensional linear model of spread of the VA supports the concept that localized fluctuation of electrical properties of damaged myocardium alters the spectral content of the activation wavefront. Expected normal conduction shows high MSC_m estimates at the start and at the end of VA. Conversely, the evidence of abnormal substratum is associated with reduction of MSC_m at both start and end of VA.

REFERENCES

[1] G. Breithardt, M. Borggrefe, A. Martinez-Rubio, “Late Potentials in Post-Infarction Period: Prognostic Significance”, In: N. El-Sherif, G. Turitto (Eds) High Resolution Electrocardiography, New York: Futura, pp. 405-425, 1992.

[2] P.R.B. Barbosa, J. Barbosa-F^o, C.A.M. De-Sá, et al. “Assessment of the Ventricular Late Potentials in HIV Positive Patients based on the RR Interval Histogram.” Computers in Cardiology; vol. 26, pp. 327‑330, 1999.

[3] A.M.L.F.M. De Sá, D.M. Simpson, A.C.F Infantosi, “Estudo da Função de Coerência Aplicada a Sinais de EEG.”[In Portuguese] Revista Brasileira de Engenharia - Caderno de Engenharia Biomédica, vol. 10, pp. 39-56, 1994.