Automatic heart sound detection in pediatric patients without electrocardiogram reference via pseudo-affine Wigner–Ville distribution and Haar wavelet lifting

https://doi.org/10.1016/j.cmpb.2013.11.018Get rights and content

Abstract

Having in mind the availability of electronic stethoscopes, phonocardiograms (PCGs) have become popular for cardiovascular functionality monitoring and signal processing applications. Detection of fundamental heart sounds (HSs), S1s and S2s, is considered to be a crucial step in PCG analysis. Electrocardiogram (ECG), noted as a reference signal, is often synchronously recorded in order to simplify the S1/S2 detection process. Nevertheless, electronic stethoscopes are frequently used without additional ECG equipment. We propose a new algorithm for automatic fundamental HSs detection via: joint time-frequency representation based on pseudo affine Wigner–Ville distribution (PAWVD), Haar wavelet lifting scheme (Haar-LS), normalized average Shannon energy (NASE) and autocorrelation. The performance of the proposed algorithm was calculated on both normal (50) and pathological (75) PCG recordings, eight seconds long each, contributed by 125 different pediatric patients. The algorithm showed relatively high recall (90.41%) and precision (96.39%) rates of S1/S2 detection procedure in a variety of PCG signals, without ECG as a reference. Furthermore, it indicated the ability to overcome splitting within the S1/S2 heart sounds.

Introduction

Phonocardiography and auscultation represent low-cost and non-invasive methods for cardiovascular system monitoring. Auscultation represents acquisition of mechanical vibrations from the body surface within the frequency range of audible sound. It is common belief that auscultation provides important information for cardiovascular diagnostic evaluation. Nowadays, electronic stethoscopes and devices for digital data acquisition of vibro-acoustic cardiosignals are commercially available. The availability at relatively low price enables widespread use of graphically recorded heart sounds (HSs) in an e-health environment [1]. The analysis of phonocardiograms (PCGs) has shown to have a large impact on auscultation decisions [2], [3], [4].

In today's health care oriented society, systems that provide monitoring cardiac and breathing sounds are rapidly developing and seem promising for the future [5], [6], [7]. They have been recognized as valuable in providing medicine and telemedicine services in patient's natural surroundings. Typical system for telemonitoring is represented in Fig. 1. The PCG and/or other signals recorded by the sensor network are transferred from the local transmitter to a distant receiver via communication network (e.g. internet). Transfer can be performed in a ‘store-and-forward’ or real-time manner.

After or during PCG acquisition, the first step in the analysis is to determine points or intervals within the recorded data that can be used as references for further analysis. Automatic or semi-automatic PCG analysis should be available locally or enabled at the receiver side, where a medical professional is present (Fig. 1). The analysis usually includes detection of S1/S2s that represent fundamental heart sounds found in the low-frequency (LF) domain, dominant from the energy standpoint.

There have been many ways to automatically analyze PCG. Most of them are oriented toward automatic detection of fundamental HSs. These sounds are considered as valuable intervals for further PCG analysis. The rhythmic (tic-tac) contraction of the heart produces systoles and diastoles which can be recorded by electrocardiogram (ECG) and/or PCG devices [3]. In ECG signal the QRS complex is highly recognizable within each heart beat and may be used as a reference for identifying the fundamental heart sounds S1 and S2 in the PCG signal. Unfortunately, such an approach needs synchronous recording of ECG and PCG. Very often synchronously recorded ECG is not available, particularly when using portable recording devices and out-of-clinic examinations, when the determination of fundamental heart sounds becomes a difficult task.

The goal of our research described in this paper is to automatically detect fundamental heart sounds within phonocardiogram without ECG as a reference. The paper is organized as follows. Section 2 gives an overview of the related work. The proposed methodology is described in Section 3. Results obtained in the study are reported in Section 4. Finally, discussion of the results and conclusions are given in Sections 5 Discussion, 6 Conclusions, respectively.

Section snippets

Related work

In the past few decades, phonocardiography accompanied with low-cost cardiovascular system analysis has emerged as an active area of research. One of the major concerns regarding PCGs in signal processing community is an adequate representation of the clinic information, where component recognition within a PCG signal probably occupies the most important role in PCG analysis-related applications [8]. There have been a lot of attempts reported for S1/S2 detection, where two main approaches can

Methodology

Block diagram for the fundamental HSs detection is depicted in Fig. 2. Proposed approach includes:

  • data preprocessing after signal acquisition,

  • window-based data JTF analysis and fundamental cardiac information determination, and

  • fundamental HSs identification based on the determined information and additional analysis (feature extraction).

Determination of fundamental cardiac information is related to dominant sound interval detection from the energy standpoint (i.e. the detection of candidates

Results

The proposed approach consists of three main stages: coarse detection, fine detection and identification of S1/S2s. In the first stage, coarse detection consisted of:

  • S1/S2 candidate detection in low-frequency domain (set D).

  • Simultaneously, potential beat duration is calculated in the starting point of each detected candidate.

The second stage is used to improve the coarse detection results by:
  • high-value peak detection via Haar-LS technique and NASE curve calculation, and

  • selection of the

Discussion

The proposed automatic S1/S2 detection algorithm is characterized by high recall and precision values (Table 3) without using any ECG equipment or other auxiliary recording besides PCG. It has been shown that:

  • R peak cannot be considered as inevitable for S1/S2 detection, contrary to beat duration.

  • PAWVD is suitable for S1/S2 candidate detection since it is based on CWT correlation.

  • The proposed methodology has overcome the splitting problem in fundamental heart sounds. A few detected fundamental

Conclusions

Acquisition of PCGs can have a large impact on improving the results of auscultation by increasing the confidence of examiners. Our research was addressed mainly to pediatric patients. Proposed S1/S2 detection algorithm is useful for stand-alone analysis. It can be used in order to avoid dependence of S1/S2 detection on ECG equipment. This does not mean that the expert's final decision regarding S1/S2 detection should be neglected. Nevertheless, a step toward automation can be made in the

Ethical approval

Acquired phonocardiograms (as a part of integrated database formed in the period of 2008–2012) were used for the testing purposes as approved by the independent ethics committee of the Health Center “Zvezdara” (No. 2140/3, 2010/05/25).

Conflicts of interest

There are no conflicts of interest.

Acknowledgements

This research is supported and partially funded by Ministry of Education, Science and Technological Development of the Republic of Serbia, as a project No. III44009. Authors are grateful to Health Center “Zvezdara”, Belgrade, Serbia, for providing medical dataset and support.

References (51)

  • C.N. Gupta et al.

    Neural network classification of homomorphic segmented heart sounds

    Applied Soft Computing

    (2007)
  • E.A. Balas et al.

    Distance technologies for patient monitoring

    British Medical Journal

    (1999)
  • M.E. Tavel

    Cardiac auscultation: a glorious past – but does it have a future?

    Circulation

    (1996)
  • A.M. Chizner

    Cardiac auscultation: rediscovering the lost art

    Current Problems in Cardiology

    (2008)
  • A. Woywodt et al.

    A novel multimedia tool to improve bedside teaching of cardiac auscultation

    Postgraduate Medical Journal

    (2004)
  • E. Kail et al.

    Internet digital phonocardiography in clinical settings and in population screening

    Computers in Cardiology

    (2004)
  • F.L. Hedayioglu et al.

    Development of a tele-stethoscope and its application in pediatric cardiology

    Indian Journal of Experimental Biology

    (2007)
  • E. Kosa et al.

    Experiences with fetal phonocardiographic telemonitoring and future possibilities

  • L. Resnekov

    Automation in cardiology

    British Heart Journal

    (1971)
  • A. Iwata et al.

    Algorithm for detecting the first and the second heart sounds by spectral tracking

    Medical & Biological Engineering & Computing

    (1980)
  • R.J. Lehner et al.

    A three-channel microcomputer system for segmentation and characterization of the phonocardiogram

    IEEE Transactions on Biomedical Engineering

    (1987)
  • M.A.R. Santos et al.

    Detection of first and second cardiac sounds based on time frequency analysis

  • M.B. Malarvili et al.

    Heart sound segmentation algorithm based on instantaneous energy of electrocardiogram

    Computers in Cardiology

    (2003)
  • C.S. Lima et al.

    Phonocardiogram segmentation by using hidden Markov models

  • A.F. Quiceno et al.

    Effective phonocardiogram segmentation using nonlinear dynamic analysis and high-frequency decomposition

    Computers in Cardiology

    (2008)
  • M.W. Groch et al.

    A new heart-sounds gating device for medical imaging

    IEEE Transactions on Biomedical Engineering

    (1992)
  • A. Gharehbaghi et al.

    An automatic tool for pediatric heart sounds segmentation

    Computing in Cardiology

    (2011)
  • G. Amit et al.

    Automatic segmentation of heart signals

  • E. McDonnell et al.

    Processing and analysis of fetal phonocardiograms

  • L. Huiying et al.

    Heart sound segmentation algorithm based on heart sound envelogram

    Computers in Cardiology

    (1997)
  • S. Omran et al.

    A heart sound segmentation and feature extraction algorithm using wavelets

  • P. Wang et al.

    First heart sound detection for phonocardiogram segmentation

  • D. Kumar et al.

    Detection of S1 and S2 heart sounds by high frequency signatures

  • J. Zhong et al.

    Automatic heart sound signal analysis with reused multi-scale wavelet transform

    International Journal of Engineering and Science

    (2013)
  • J. Vepa et al.

    Segmentation of heart sounds using simplicity features and timing information

  • Cited by (32)

    • DA-Bi-SRU for water quality prediction in smart mariculture

      2022, Computers and Electronics in Agriculture
      Citation Excerpt :

      The noise reduction methods mainly include: moving average method, smoothing method, basis function least squares method, quadratic bilinear time–frequency analysis, Fourier transform and wavelet transform, etc. Among them, wavelet analysis (Chen, 2016; Pan et al., 2007; Parmar and Bhardwaj, 2013; Gavrovska et al., 2014; Wickersham et al., 2014; Rui-zhi, 2013) can describe the non-stationarity of dynamic signals, extract local information in the frequency domain and time domain of the signal at the same time, and achieve the separation and extraction of the frequency characteristics of the water quality signals. It has the characteristics of multi-scale analysis and “mathematical microscopy” (Li, 2013), which is a tool to achieve efficient and convenient noise reduction and data cleaning for water quality.

    • Automatic analysis of the heart sound signal to build smart healthcare system

      2022, Internet of Multimedia Things (IoMT): Techniques and Applications
    • Ensemble of kernel extreme learning machine based random forest classifiers for automatic heartbeat classification

      2021, Biomedical Signal Processing and Control
      Citation Excerpt :

      The non-stationary and nonlinear characteristics of the ECG waveforms are the main challenges faced by the time-domain and the frequency-domain feature extraction approaches. Then, time-frequency techniques appeared as a solution, such as Short-time Fourier transform [22], Wigner-Ville distribution [23], empirical mode decomposition [24] and wavelet transform. The wavelet transform methods are especially widely explored [25–28] owing to their good performance in analyzing non-stationary signals.

    • Information hiding methods for E-Healthcare

      2020, Smart Health
      Citation Excerpt :

      The QRS complex is the combination of the Q wave, R wave, and S wave and represents ventricular depolarization. After detecting R waves (Bacharova, Bang, Szathmary, & Mateasik, 2014), Haar lifting wavelet transform (Gavrovska, Bogdanović, Reljin, & Reljin, 2014) is applied again on the ECG signal. Then, the non-QRS high frequency wavelet coefficients are selected by comparing and applying index subscript mapping.

    • Classification of heart sound signal using curve fitting and fractal dimension

      2018, Biomedical Signal Processing and Control
      Citation Excerpt :

      Segmentation using ECG signal have problems. PCG and ECG recording is not available in many cases, especially when the portable recording devices are used outside the clinic [20]. ECG signals are sometimes impaired by some heart diseases, such as ventricular hypertrophy.

    • Study of hyperparameters in the semantic segmentation of heart sounds

      2023, Proceedings of the 19th International Symposium on Medical Information Processing and Analysis, SIPAIM 2023
    View all citing articles on Scopus
    View full text