f
Performance assessment of standard algorithms for dynamic R-T interval measurement: comparison between R-Tapex and R-Tend approach A. Porta 1
G. Baselli 2 F. L o m b a r d i 3 S. C e r u t t i I M . Del G r e c o 4 F. Ravelli 4 G. N o l l o 4
R. A n t o l i n i 4
1Dipartimento di Bioingegneria, Politecnico di Milano, Milano, Italia 2Dipartimento di Elettronica per I'Automazione, Universita' di Brescia, Brescia, Italia 3Medicina Interna II, Ospedale L. Sacco, Universita' di Milano, Milano, Italia 4Centro Materiali e Biofisica Medica - ITC, via Sommarive 18, 38050 Povo, Trento, Italia
AbstractEThree automatic approaches to ventricu/ar repolarisation duration measurement (F:~---Tapex, R--Ten d threshold and R-Tend fining methods) are compared on computergenerated and real ECG signals, in relation to their reliability in the presence of the most common electrocardiographic artefacts (i.e. additive broadband noise and additive and multiplicative periodical disturbances). Simulations permit the evaluation of the amount of R-T beat-to-beat variability induced by the artefacts. The R--Tend threshold method performs better than the R--Tend tining one, and, hence, the latter should be used with caution when R--Tend variability is addressed. Whereas the R-Tepex method is more robust with regard to broadband noise than the R--Ten d threshold one, the reverse situation is observed in the presence of periodical amplitude modulations. A high level of broadband noise does not prevent the detection of the central frequency of underlying R-T periodical changes. Comparison between the power spectra of the beat-to-beat R-T variability series obtained from three orthogonal ECG leads (X,Y,Z) is used to assess the amount of real and artefactual variability in 13 normal subjects at rest. The R-Tapex series displays rhythms at high frequency (HF) with a percentage power on the Z lead (57.1 • greater than that on the X and Y leads (41.9• and 46.1 • respectively), probably because of respiratory-related artefacts affecting the Z lead more remarkably. More uniform HF power distributions over X,Y,Z leads are observed in the R--Tend threshold series (31.8 • 39.2• and 35.1 • respectively), thus suggesting minor sensitivity of the R--Tend threshold measure to respiratory-related artefacts. Keywords---Q--T interval, Ventricular repolarisation duration, Automatic R - T measurement, R-T variability, Time- and frequency-domain analysis Med. Biol. Eng. Comput., 1998, 36, 35-42
1 Introduction
THE Q-T interval, defined as the temporal distance between the Q-wave onset and the T-wave offset, represents an indirect measure of the ventricular repolarisation (VR) duration on surface electrocardiograms. The study of the Q-T period is an issue of growing interest, especially owing to the clinical implications of Q-T prolongation and dispersion. The lengthening of the Q-T interval, and/or of the frequency-corrected Q-T (Q-To) has been considered to be predictive of severe arrhythmic events in the long Q-T syndrome (SCHWARTZet el., 1975), in patients after myocardial infarction (SCHWARTZ and WOLF, 1978) and even in an apparently healthy population (SCHOUTEN et al., 1991). The Q-T dispersion indexes have
Correspondenceshould be addressed to Dr. NoIIo; email:
[email protected] First received 12 July 1996and in final form 5 June 1997 9 IFMBE:1998 Medical & Biological Engineering & Computing
been closely related to an increase in the risk of sudden death and malignant arrhythmias (BARRet aI., 1994; DAY et al., 1990). The above applications are based on static measures often obtained on a median pattern extracted from a few consecutive beats; nonetheless, the VR process is dynamic and varies from beat to beat, mainly as a function of changes in the preceding R-R intervals (BAZETT, 1920; ARNOLD et al., 1982; M~SON BLANCHE et al., 1996) and of many other factors, including autonomic nerve tone, presence of disease and pharmacological interventions (BROWNE et al., 1983; HIGHAM and CAMPBELL, 1994; EDVARDSSONand OLSSON, 1981). Therefore beatto-beat measurement of the Q-T interval has been proposed as a means to gain insight into the spontaneous dynamic variations of the VR process. Recently, analysis in the frequency domain of indexes of VR duration has proved that the beat-to-beat series of consecutive R-T intervals exhibits the same oscillations found in the R-R series, referred to as low-frequency (LF) (around 0.1 Hz) and high-frequency (HF) (synchronous with respira-
January 1998
35
tory activity) rhythms (NOLLO et aI., 1992; MERRI et al., 1993; LOMBARDI et al., 1996). These studies, which hypothesise a relationship between the power of the oscillations of the R-T measure and the sympatho-vagal balance at ventricular level, enlarged the potential applications of parameters describing the VR phenomenon. Two alternative intervals have been proposed as indexes of the VR duration. These measures were defined as the temporal distance between the peak of the QRS complex and the Twave apex (R-Tapex) and between the R-peak and the T-wave end (R-Tend). Although there is general agreement on the use of parabolic interpolation to minimise the jitters in the detection of the T-wave apex (MERRI et al., 1993; PORTA et al., 1994), no standard, uniform method is used to identify the Twave end. Some groups define the T-wave end as where the first derivative of the T-wave downslope falls below a threshold (LAGUNA et al., 1990; SPERaNZA et al., 1993). In other works, the T-wave offset is detected by considering the intersection between the iso-electric line and the tangent at the steepest point of the T-wave downslope (LEPESCHKINand SURAWICZ, 1952; YAMADA et al., 1993). Although the effect of disturbances such as broadband noise due to muscle activity, slow baseline wandering, cardiac axis movements and amplitude modulations related to respiration on the Twave fiducial point detection is evident (PORTA et al., 1994), no exhaustive comparison of the robustness of different methods has yet been carried out. The aim of this paper is to compare the main algorithms for dynamic R-T measurement with regard to their ability to give reliable results in the presence of the usual disturbances affecting ECG signals. The effect of artefacts on R-T measurement accuracy was evaluated by simulating the most common electrocardiographic artefacts on a computergenerated ECG signal designed to have constant R-Tapex and R-T~d periods (i.e. no R-T variability). The accuracy of the R-T measurement procedures in detecting periodic R-T variability was assessed by adding a different level of noise to another artificial ECG signal arranged to have a pre-defined rhythmical R-T variability at 0.1 and 0.3 Hz. Finally, the R-Tap~x and R-T~,d variability series were extracted from real data recorded from healthy subjects at rest on X,Y,Z orthogonal leads. Parameters both in the time and frequency domains were calculated and compared to assess whether different information could be obtained from different ECG leads and R-T measurement techniques.
2.2 Automatic R - T measurement methods All the implemented automatic R-T measurement procedures were based on QRS complex detection and subsequent baseline removal. The QRS recognition and R-wave maximum fiducial point detection were obtained by means of a derivative-threshold algorithm and parabolic fitting. The oscillation of the iso-electric line was estimated by means of cubic spline interpolation, based on 11 iso-electric points detected on 11 beats (five before and five after the current one). Each point occurred 40 ms before the onset of the QRS complex. To detect the T-wave apex fiducial point, the T-wave maximum (minimum) was searched for in a time window ranging between 0.15 and 0.4 times the preceding R-R interval. As shown in Fig. Ia, to reduce the effect of noise in determining the T-wave apex, a parabolic fitting was performed around the T-wave maximum (MERRI et al., 1993; PORTA et al., 1994), thus obtaining the R-Tapex interval duration. Two well-known algorithms for automatic measurement of the T-wave offset were analysed. The first approach (LAGUNA et al., 1990; SPERANZAet al., 1993) identified the T-wave end with the point where the absolute value of the first derivative of the T-wave downslope became smaller than a threshold proportional to the T-wave derivative maximum absolute
Rma~x
I
