ARTICLE

QRS Complex Wavelet Analysis Can Distinguish Patients with and without Heart Failure, in the Presence of Left Bundle Branch Block

A B S T R A C T

Background: Left bundle branch block (LBBB) in heart failure (HF) patients is a negative predictor of survival. This pattern is occasionally recorded in individuals without structural heart disease. The LBBB morphology has not been previously analyzed in a time-frequency domain using wavelet analysis), and thus the factors distinguish LBBB patients from individuals without structural heart disease remain unexplored. The purpose of this analysis was to investigate the variations and the differences in LBBB morphology between healthy individuals with LBBB and patients with HF and LBBB.
Methods: HF patients with LBBB and individuals with LBBB were included in this study. Signal-averaged 90-second Holter monitor recordings were extracted from each subject in orthogonal leads. QRS decomposition in 9 time-frequency bands (TFB) was performed using Complex Morlet wavelets transformation, while the mean and maximum energies of the QRS complexes were calculated for each of the 9 TFBs. The wavelet parameters of HF patients were compared with those of healthy controls.
Results: Wavelet analysis was performed on ECG recordings of 69 HF patients and 17 individuals without cardiac disease. The mean and max wavelet energies of the QRS complex in all TFBs were higher for heart failure patients with LBBB, as compared to healthy individuals with LBBB. Differences were statistically significant in TFB4 and TFB7 (max energy, axis X), TFB4 and TFB7 (max energy, axis Y) and TFB4 and TFB7 (mean energy, axis Y). A multivariate logistic regression model, comprising of the aforementioned wavelet parameters, proved reasonably capable of distinguishing between HF patients and healthy controls with LBBB (AUC=0.854, 80.2% sensitivity and 80.3% specificity).
Conclusion: QRS wavelet analysis revealed differences in the template of the QRS complex between healthy individuals with LBBB and heart failure patients with LBBB. This feature could be used as part of the diagnostic algorithm, a possibility that should be investigated further.

Keywords

Left bundle branch block, signal processing, heart failure, QRS wavelet analysis

Introduction

Left bundle branch block (LBBB) is reflected on the electrocardiogram (ECG) as a QRS prolongation above 120ms, a delay of intrinsic deflection in leads V5 and V6 of more than 60ms and an absence of Q waves in leads I, V5 and V6 [1]. On the basis of additional insights from computer simulations, the investigators propose stricter criteria for complete LBBB that include a QRS duration ≥140 ms for men and ≥130 ms for women, along with mid-QRS notching or slurring in ≥2 contiguous leads [2]. It is characterized by obstruction of the depolarization’s signal through the left ventricle’s conduction system. LBBB is usually present in patients with cardiac disease, most commonly coronary artery disease (CAD) and chronic heart failure (CHF). LBBB occurs in up to 30% of patients with heart failure [3]. In these patients, LBBB is an ominous finding, as its presence correlates with increased mortality and frequency of adverse events [4, 5]. Nonetheless, LBBB is occasionally encountered in healthy individuals, having a prevalence of less than 1% in the general population [6]. Even in these cases, LBBB confers such individuals with a mortality risk of 1.3 [6].

LBBB is associated with poor electromechanical function of the left ventricle, as regional delays in depolarization cause dyssynchronous activation of the ventricular walls. Reduced systolic shortening in regions of the myocardium that depolarize early during systole has been observed as the cause of poorer LV function in the setting of LBBB [7]. Cardiac resynchronization therapy (CRT) is performed by pacing both ventricles and is very effective at ceasing and even undoing the deleterious effects of LBBB on cardiac function and remodeling [8]. In fact, the presence of LBBB predicts which HF patients with reduced ejection fraction stand to benefit the most from CRT therapy [9].

In addition to QRS complex duration and morphology analysis, our group was also subject to a Complex Morlet wavelets analysis (‘cmor’) of QRS complex, which constitutes a novel technique of orthogonal ECG analysis based on the combination of time-domain and frequency- domain [10]. This technique has been found useful in the detection of small signal components in large ECG waves [10, 11].

Surface ECG provides a time-domain and a frequency-domain analysis of the electrical activation of the heart. However, the frequency content of the signal may provide additional information. The wavelet transform is a mathematical function that has been used for almost three decades as an alternative to the traditional time-domain methods providing a time- frequency domain analysis [12, 13]. Wavelet decomposition of the signal-averaged electrocardiogram has been proposed as a method of detecting small and transient irregularities hidden within the QRS complex with marked accuracy and reproducibility [14, 15].

Wavelet transform (WT) is a widely used method for time frequency transformations and it was adopted by several studies in order to capture the time specific frequency content appearing in the biomedical signals such as ECG [16]. Wavelet analysis of QRS morphologies has been performed in the past, enabling the accurate detection of the Q fiducial point [17, 18]. Also, wavelet QRS analysis has been used for the prediction of response to cardiac resynchronization therapy [19]. At the same time, beat-to-beat wavelet analysis of the P-wave has uncovered significant differences in wavelet properties between healthy controls and patients with paroxysmal atrial fibrillation [20].

The aim of the present study is to delineate the wavelet properties of the QRS complex in HF patients with LBBB and otherwise healthy controls that exhibited LBBB, as well as compare the findings from both groups.

Materials and Methods

I Study Population

The study was designed and executed according to the Declaration of Helsinki. After approval of our institution’s Medical Ethics Committee, all participants were provided with a consent form that they agreed to sign. HF patients with LBBB, treated in our Department, have been invited to participate in this study. To avoid heterogeneity, only patients with reduced ejection fraction (HFrEF) included in the study. The control group consists of patients treated by our hospital for non-cardiac causes and exhibited LBBB without organic heart disease. All the patients of the control group underwent a transthoracic echocardiogram and a stress test (stress- echocardiogram or treadmill test). None of them had history of hypertension and diabetes. The diagnosis of LBBB was utilized with the established ECG criteria. QRS duration (QRS>130msec for women and QRS>140msec for men) were used for this study [2]. Patients who were ventricularly paced were excluded from the study.

II ECG Recordings

ECG signals were recorded with a 3-channel digital Holter monitor (Galix Biomedical Instrumentation) for 10 minutes at rest in the X, Y and Z orthogonal leads [21]. The consistent rectangular coordinate system is illustrated in (Figure 1). The sampling frequency was 1000Hz, and the duration of each signal was 90 seconds in order to ensure an adequate and artifact-free signal of the same length for all patients.

III ECG Signal Processing and Wavelet Analysis

A signal averaging algorithm was developed to enhance the QRS signal and reduce the noise level prior to wavelet analysis. The first step was the extraction of raw data from three orthogonal leads (X,Y,Z) from the Holter monitor to the computer, followed the signal processing. The signal processing includes; the baseline wander removal which was then achieved through amplitude normalization, baseline correction and application of denoising wavelet filter [22]. Noise was filtered out through application of a de-noising wavelet filter. The linear trend from ECG and the power line interference (50Hz) were removed from every lead. Then a high pass filter and a low pass filter (f1-f2: 0.8Hz-200Hz) was applied. A one-dimensional de-noising function was applied for the wavelet filter. Also, the threshold selection rule is a heuristic variant of the first option using a soft thresholding with a symplet complex wavelet and 5 levels of the wavelet decomposition. Selected the multiplicative threshold for rescaling done using level-dependent estimation of level noise [23].

The following step was included the localization of the R points and the calculation of QRS complexes [24]. All QRS-complexes in each signal were calculated and manually inspected by two blinded in order to exclude the ectopic beats. QRS complexes in every signal were evaluated, and those found to contain noise or constitute premature ventricular contractions (PVCs) were excluded from the analysis.

Wavelet Analysis and Quantification

Wavelet transformation of the QRS complex was conducted, using the Complex Morlet wavelets analysis [25]. The outcome of a continuous wavelet transform (CWT) is a function of two variables, time and scale. Changing the scaling parameter, results in a function with different frequency content and time spanning. In this respect, each scaled version of the mother wavelet can be mapped to a function with a specific central frequency [20]. The bandwidth and the center frequency were 3 and 1 respectively.

QRS complex was divided into 3 equal parts with respect to time and 3 frequency bands, creating a total of 9 time-frequency bands (TFB). The high frequency range (150-200Hz) is found in bands TFB1, TFB2, and TFB 3; the medium frequency range (100-150Hz) is found in bands TFB4, TFB5, and TFB6; and the low frequency range: (50-150Hz) is found in bands TFB7, TFB8 and, TFB9 (Figure 2). The initial, middle and ending part of the QRS was segmented accordingly. The wavelet features that were calculated were the mean and maximum wavelet energies of the QRS complexes. These were calculated in each of the 9 time-frequency bands for each of the 3 axes, yielding 27 features per recording [26].

Comparing the Wavelet Properties between HF Patients and the Control Group

The mean and maximum wavelet energies for each time-frequency band were compared between the HF group and the controls. Mean corresponding to the mean energy of QRS complex in a specific time-frequency area and maximum corresponding to the maximum energy of QRS complex in a specific time-frequency area. In order to investigate the ability of QRS wavelet analysis to distinguish between HF patients and healthy controls, a multivariate logistic regression was applied. All wavelet parameters found to be statistically significantly different between the two groups, at a level of p<0.05, were included in this model. The following equation describes our model.

$$\log{\left[\frac{\mathrm{prob}\left(\mathrm{HF}\right)}{\mathrm{(1\ -\ prob(HF))}}\right]}\mathrm{=} \ TFB4-MaxX+\beta TFB7-MaxX+\beta TFB4-MaxY+
$$ $$\beta TFB7-MaxY+\beta TFB4-MeanY+\beta TFB7-MeanY+\varepsilon$$

The capability of this model to separate HF patients from healthy controls was tested with the ROC curve and AUC analysis. All statistical analyses were performed using IBM SPSS Statistics version 23.0 (Chicago, IL, USA). Mean and range of minimum and maximum values were used to express normally distributed variables, and median and interquartile range for non-normally distributed variables (i.e., all wavelet parameters). Differences between groups were explored with Student's t-test or Mann Whitney U test for normally and abnormally distributed variables, respectively. All p- values were considered significant at the 5% level.

Results

The total population of the study consisted of 86 subjects, 69 HF patients (39 with dilated and 30 with ischemic cardiomyopathy respectively) (Group A) and 17 otherwise healthy individuals with LBBB (Group B). Demographic data of both groups are provided in (Table 1). Wavelet parameters in each of the three orthogonal leads that appeared to differ between HF patients and healthy controls are presented in (Table 2).

Comparing the quantified wavelet properties, maximal wavelet energies of the QRS complex in all frequency bands were higher, on average, for individuals with heart failure, and statistically significant differences were found in the maximal observed wavelet energy in TFB4 (p=0.004), TFB7 (p=0.013) in axis X; TFB4 (p=0.018) and TFB7 (p=0.008) in axis Y. The mean wavelet energy was significantly increased in HF patients in TFB4 (p=0.038) and TFB7 (p=0.032) (Figure 2).

Figure 3 illustrates the wavelets of a typical HF patient and a healthy individual with LBBB. A multivariate logistic regression has been applied to detect HF by means of wavelet properties (Table 3). The area under the ROC curve was significant (AUC: 0.854; p=0.000; Figure 4, Table 4). This model achieved 80.2% sensitivity and 80.3% specificity in detecting LBBB individuals with heart failure.

Discussion

The present study was conducted in order to investigate possible differences in the QRS complex in patients with LBBB and HF, compared to healthy controls with LBBB, using a wavelet analysis of the QRS complex.

Our results confirm that there are signal components inside the QRS complex that could distinguish LBBB appearing in individuals without structural heart disease and LBBB in patients with HF and reduced EF. In particular, HF patients exhibited significantly higher maximal wavelet energies than individuals without heart diseases. Four (4) out of the six (6) wavelet parameters incorporated in the multivariate logistic regression model displayed reasonable capacity to distinguish between HF patients and healthy controls. The model was able to detect HF with a sensitivity of 80.2% and a specificity of 80.3% in the sample of the study. It would be interesting to see whether models based on wavelet properties retain their predictive capacity in larger and more heterogeneous sample sizes. It can be observed that with respect to the temporal segmentation, the features from the first part of QRS seem to be more informative than the last part of QRS. The most important features of the model correspond to the low and medium frequency band of QRS. This finding suggests conduction differences, and abnormalities related to delayed or fragmented conduction components, which is a recognizable phenomenon in LBBB, but also and even more so in ischemic hearts.

Some observations can be made about the characteristics of both groups. The mean age (65.7y for the HF group vs 67.9y for the healthy control group) and the male gender (78.3% for the HF group vs 70,6% for the control group) were not significantly differ. QRS duration was also higher in the HF group, but this can be factually attributed to the higher ventricular sizes observed in HFrEF patients, which are tied to an increase in the size of the QRS complex [27].

Conclusion

Our preliminary results indicate that time-frequency analysis of the LBBB-type QRS complex revealed significant differences in HF patients and individuals without heart diseases. The regression model which was created based on these differences achieved reasonable sensitivity and specificity in distinguishing patients with or without heart failure presenting with LBBB. QRS wavelet analysis for diagnostic purposes is still a nascent field, and more research is needed to reliably prove its capacity to distinguish HF patients with LBBB from individuals with LBBB without organic heart disease.

Acknowledgements

None.

Funding

This research was supported by the Greek State Scholarship Foundation.

Conflicts of Interest

None.

ARTICLE INFO

Download PDF View PDF

Article Info

Article Type

Research Article

Publication history

Received:

Accepted:

Published:

Copyright

© 2023 Kalliopi Papathoma. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Hosting by Science Repository.

DOI: 10.31487/j.JICOA.2020.03.06

AUTHOR INFO

Author Info

Anastasios Tsarouchas

Dimitrios Mouselimis

Efstratios Theofilogiannakos

Eleni Christaina

Ioanna Chouvarda

Kalliopi Papathoma

Nikolaos Maglaveras

Stavros Chatzimiltiadis

Vassilios Vassilikos

Corresponding Author

Kalliopi Papathoma
Laboratory of Computing, Medical Informatics and Biomedical Imaging Technologies, School of Medicine, Aristotle University, Thessaloniki, Greece

FIGURES & TABLES

REFERENCE

1. Breithardt G, Breithardt OA (2012) Left Bundle Branch Block, an Old-New Entity. J Cardiovasc Transl Res 5: 107-116. [Crossref]
2. Strauss DG, Selvester RH, Wagner GS (2011) Defining Left Bundle Branch Block in the Era of Cardiac Resynchronization Therapy. Am J Cardiol 107: 927-934. [Crossref]
3. Abraham WT, Fisher WG, Smith AL, Delurgio DB, Leon AR et al. (2002) Cardiac Resynchronization in Chronic Heart Failure. N Engl J Med 346: 1845-1853. [Crossref]
4. Tabrizi F, Englund A, Rosenqvist M, Wallentin L, Stenestrand U (2007Influence of Left Bundle Branch Block on Long-Term Mortality in a Population With Heart Failure. Eur Heart J 28: 2449-2455. [Crossref]
5. Al Daydamony MM, Mustafa TM (2017) The Relation Between Coronary Artery Disease Severity and Fragmented QRS Complex in Patients With Left Bundle Branch Block. Egypt Hear J 69: 119-126. [Crossref]
6. Francia P, Balla C, Paneni F, Volpe M (2007) Left Bundle-Branch Block--Pathophysiology, Prognosis, and Clinical Management. Clin Cardiol 30: 110-115. [Crossref]
7. Nguyên UC, Verzaal NJ, van Nieuwenhoven FA, Vernooy K, Prinzen FW (2018) Pathobiology of Cardiac Dyssynchrony and Resynchronization Therapy. Europace 20: 1898-1909. [Crossref]
8. Goldenberg I, Kutyifa V, Klein HU, Cannom DS, Brown MW et al. (2014) Survival with Cardiac-Resynchronization Therapy in Mild Heart Failure. N Engl J Med 370: 1694-1701. [Crossref]
9. Zareba W, Klein H, Cygankiewicz I, Hall WJ, McNitt S et al. (2011) Effectiveness of Cardiac Resynchronization Therapy by QRS Morphology in the Multicenter Automatic Defibrillator Implantation Trial-Cardiac Resynchronization Therapy (MADIT-CRT). Circulation 123: 1061-1072. [Crossref]
10. Vassilikos V, Dakos G, Chatzizisis YS, Chouvarda I, Karvounis C et al. (2011) Novel Non-Invasive P Wave Analysis for the Prediction of Paroxysmal Atrial Fibrillation Recurrences in Patients Without Structural Heart Disease: A Prospective Pilot Study. Int J Cardiol 153: 165-172. [Crossref]
11. de Oliveira HM, de Souza DF (2006) Wavelet analysis as an information processing technique. IEEE.
12. Figliola A, Serrano E (1997) Analysis of Physiological Time Series Using Wavelet Transforms. IEEE Eng Med Biol Mag 16: 74-79. [Crossref]
13. Morlet D, Peyrin F, Desseigne P, Touboul P, Rubel P (1993) Wavelet Analysis of High-Resolution Signal-Averaged ECGs in Postinfarction Patients. J Electrocardiol 26: 311- 320. [Crossref]
14. Yi G, Hnatkova K, Mahon NG, Keeling PJ, Reardon M et al. (2000) Predictive Value of Wavelet Decomposition of the Signal-Averaged Electrocardiogram in Idiopathic Dilated Cardiomyopathy. Eur Heart J 21: 1015-1022. [Crossref]
15. Rubel P, Hamidi S, Behlouli H, Couderc JP, Fayn J et al. (1996) Are Serial Holter QT, Late Potential, and Wavelet Measurement Clinically Useful?. J Electrocardiol 29: 52- 61. [Crossref]
16. Langberg JJ, Burnette JC, McTeague KK (1998) Spectral Analysis of the Electrocardiogram Predicts Recurrence of Atrial Fibrillation After Cardioversion. J Electrocardiol 31: 80-84. [Crossref]
17. Park JS, Lee SW, Park U (2017) R Peak Detection Method Using Wavelet Transform and Modified Shannon Energy Envelope. J Health Eng 2017: 4901017. [Crossref]
18. Jenkal W, Latif R, Toumanari A, Dliou A, El B’Charri O (2016) Enhanced algorithm for QRS detection using discrete wavelet transform (DWT). Proc Int Conf Microelectron ICM 39-42.
19. Vassilikos VP, Mantziari L, Dakos G, Kamperidis V, Chouvarda I et al. (2014) QRS Analysis Using Wavelet Transformation for the Prediction of Response to Cardiac Resynchronization Therapy: A Prospective Pilot Study. J Electrocardiol 47: 59-65. [Crossref]
20. Filos D, Chouvarda I, Tachmatzidis D, Vassilikos V, Maglaveras N (2017) Beat-to-beat P-wave Morphology as a Predictor of Paroxysmal Atrial Fibrillation. Comput Methods Programs Biomed 151: 111-121. [Crossref]
21. Frank E (1956) An Accurate, Clinically Practical System for Spatial Vectorcardiography. Circulation 13: 737-749. [Crossref]
22. Donoho DL (1994) De-noising by soft-thresholding. IEEE 41: 613- 627.
23. Chouvarda I, Maglaveras N, Pappas C, Boufidou A (1998) An Integrated Environment for ECG Processing. Stud Health Technol Inform 52: 986-989. [Crossref]
24. Pan I, Tompkins WJ (1985) A Real-Time QRS Detection Algorithm. IEEE Trans Biomed Eng 32: 230-236. [Crossref]
25. Morlet D, Peyrin F, Desseigne P, Touboul P, Rubel P (1991) Time-scale analysis of high-resolution signal-averaged surface ECG using wavelet transformation. Proceed Comput Cardiol 393-396.
26. Vassilikos V, Dakos G, Chatzizisis YS, Chouvarda I, Karvounis C et al. (2011) Novel Non-Invasive P Wave Analysis for the Prediction of Paroxysmal Atrial Fibrillation Recurrences in Patients Without Structural Heart Disease: A Prospective Pilot Study. Int J Cardiol 153: 165-172. [Crossref]
27. Stewart RA, Young AA, Anderson C, Teo KK, Jennings G et al. (2011 Relationship Between QRS Duration and Left Ventricular Mass and Volume in Patients at High Cardiovascular Risk. Heart 97: 1766-1770. [Crossref]