Wheeze detection based on time-frequency analysis of breath sounds

https://doi.org/10.1016/j.compbiomed.2006.09.007Get rights and content

Abstract

Abnormal breath sounds like wheezes are observed in patients with obstructive pulmonary diseases. The aim of this study was to construct an automatic technique for wheeze detection and monitoring using spectral analysis. Wheezes from 13 patients with diagnosed asthma, chronic obstructive pulmonary disease and pneumonia were recorded and a time-frequency wheeze detector (TF-WD) based on TF wheeze characteristics was constructed. The TF-WD was evaluated using 337 wheezes by comparing its findings with those from clinical auscultation performed by two experts. In addition, the TF-WD was tested against artificial noise. The experimental and testing results justified the efficient performance and high noise robustness of the TF-WD.

Introduction

From Laënnec's era [1] until nowadays auscultation has been adopted by the physicians as an easy, fast and noninvasive way to evaluate and diagnose patients with lung diseases. Nevertheless, auscultation exhibits noticeable drawbacks, as it suffers from subjectivity and variability in the interpretation of its diagnostic information. This diagnostic value can be better revealed when sound signal digitization and processing techniques are employed [2], [3]. In that way, novel diagnostic tools that objectively track the characteristics of the relevant pathology and assist the clinician in everyday practice could be introduced.

Wheezes are continuous adventitious lung sounds that are commonly observed in patients with obstructive airways diseases, such as asthma (AS) and chronic obstructive pulmonary disease (COPD) [4]. As Pasterkamp et al. [5] note, wheeze has been extensively used as an indicator of airway obstruction in infants, as a parameter to gauge the severity of AS, or as a classifier in epidemiologic surveys. Moreover, Kercsmar [6] refers to wheezes as the acoustic manifestation of airways obstruction. In addition, Bentur et al. [7] employ wheeze monitoring for the assessment of nocturnal AS and response to therapy. Nevertheless, ‘wheeze’ is not a sharply defined concept and many definitions, both as verbatim and quantitative descriptions, have been proposed in the literature [8]. In this study, wheezes are considered to be breath sounds with time duration greater than 150 ms [2]. Due to this duration, wheezes are differentiated from other abnormal sounds, such as crackles, which typically last less than 20 ms [9]. The term adventitious implies that they are superimposed on normal breath sounds. The waveform of a wheeze in time domain resembles that of a sinusoidal sound, justifying its musicality; hence, exhibiting distinct peaks in the frequency domain (>100Hz) [9].

Although simple hearing easily recognizes the musical sound of a wheeze, computerized analysis of lung sounds allows a reproducible quantification of wheezing, contrary to subjective auscultation [10]. The most straightforward methods for automatic wheezing detection are based on searching for peaks in successive spectra [5]. Efforts to further improve their reliability combined spectra with amplitude, duration and pitch range of wheezes criteria or rules [10], [11], [12], [13]. However, these algorithms show a great correlation of the number of detected wheezes with sound signal amplitude. Efforts have been made towards the design of wheeze detectors that are independent of sound attenuation, yet, focusing only in a segment of forced exhalation (1.2–0.2 l/s) [14].

In the present study, an enhanced wheeze detector is introduced, namely time-frequency wheeze detector (TF-WD), which automatically locates and identifies wheezing-episodes during breath sound recordings based on time-frequency (TF) analysis of the breath sound signal. The TF-WD was evaluated on wheeze-data recorded from patients with AS, COPD and pneumonia (PN). The acquired wheezes were of varied amplitude and occurred at different time instances of the breathing cycle. Moreover, the noise insusceptibility of the TF-WD was tested on wheezes corrupted with artificial noise that resembled other breath sounds and artifacts at different signal-to-noise ratios (SNRs), proving increased performance of the TF-WD when compared to previous works.

Section snippets

Study design

This was a pilot study to test, in practice, the efficiency of the proposed TF-WD technique. All recording procedures took place at Philipps University of Marburg, Germany, after obtaining the approval of its ethical committee. In addition, each recording procedure was explained to the subjects and all subjects signed a detailed informed consent form.

Study dataset

Breath sound signals from 13 wheezing patients were used as the dataset of the present study. These signals, drawn from the Marburg Respiratory

Experimental results

Descriptive experimental results from the analysis of the sound dataset derived from the cases of Table 1 are depicted in Fig. 2. In all subfigures, i.e., Figs. 2(i)–(iii), (a)–(c) correspond to the recorded sound signal (with the normalized airflow superimposed with a dotted-line), its representation in the TF domain, and the corresponding TF output of the TF-WD algorithm, respectively. In particular, Fig. 2(i) refers to a case with low noise contamination (Table 1, case 9) where two wheezes

Discussion

This study demonstrates that TF-based analysis of the recorded breath sounds is an efficient and robust method for computerized wheeze detection. Since wheezes are often present in sound recordings from patients with airways obstruction [8], the TF-WD technique could constitute a practical non-invasive tool for monitoring related pathologies, such as AS, COPD and PN.

The adopted sampling frequency (5512 Hz) allowed us to preserve signal characteristics up to 2756 Hz. Nevertheless, our study was

Conclusion

An efficient wheeze detector was presented in this study that could be successfully used in clinical practice due to its noise robustness, fast implementation and increased accuracy. Performance evaluation through experimental and testing results revealed increased detectability, sensitivity and specificity of the proposed method for all analyzed cases included in the study group. Further extension of the latter, either in the number of subjects and/or pathologies, will constitute a focus of

Summary

In this study we have performed time-frequency (TF) analysis of the breath sounds recorded from patients with obstructive airways diseases, such as AS, COPD and pneumonia. The TF analysis focused in the detection of wheezes in the recordings, forming a TF-based wheeze detector (TF-WD). The aim of the study is, apart from the introduction of the TF-WD scheme, the examination of its performance in real data under different levels of real and artificial noise contamination.

A number of 422 wheezes

Acknowledgments

The authors would like to thank Prof. Thomas Penzel and Dr. Volker Gross from Philipps University of Marburg, Germany, for carrying out the sound recordings, qualitatively evaluating them and providing fruitful comments that greatly contributed to the improvement of this manuscript.

Styliani A. Taplidou was born in Thessaloniki, Greece in 1980. She received her Diploma degree in Electrical and Computer Engineering in 2003 from the Department of Electrical and Computer Engineering at the Aristotle University of Thessaloniki, Thessaloniki, Greece. She is currently pursuing her Ph.D. on biacoustic signals analysis at the same department. Her research interests include wavelet analysis, higher-order statistics, modeling and digital signal processing applications. She is a

References (32)

  • J.E. Earis et al.

    European respiratory society task force report: computerised respiratory sound analysis (CORSA): recommended standards for terms and techniques

    Eur. Respir. Rev.

    (2000)
  • A.R.A. Sovijärvi et al.

    Characteristics of breath sounds and adventitious respiratory sounds

    Eur. Respir. Rev.

    (2000)
  • G. Charbonneau et al.

    Basic techniques for respiratory sound analysis

    Eur. Respir. Rev.

    (2000)
  • Y. Shabtai-Musih et al.

    Spectral content of forced expiratory wheezes during air, He, and SF6 breathing in normal humans

    J. Appl. Physiol.

    (1992)
  • S.A. Taplidou, L.J. Hadjileontiadis, T. Penzel, V. Gross, S.M. Panas, WED: an efficient wheezing-episode detector based...
  • Y. Qiu et al.

    Automatic wheeze detection based on auditory modelling

    Proc. I MechE Part H J. Eng. Med.

    (2005)
  • Cited by (188)

    • Automated respiratory sound analysis

      2022, Wearable Sensing and Intelligent Data Analysis for Respiratory Management
    • An incremental algorithm based on multichannel non-negative matrix partial co-factorization for ambient denoising in auscultation

      2021, Applied Acoustics
      Citation Excerpt :

      Although today there are more advanced medical tools to analyze the status of the heart and lung, such as chest radiography, electrocardiography (ECG), spirometry or laboratory analyses, auscultation is still one of the most widely used techniques to detect any cardiac or pulmonary disease. However, the diagnosis derived from auscultation shows two main limitations: i) high subjectivity due to the physician’s expertise to recognize sounds that reveal any physiological disorder [3]; ii) high dependence on the ambient noise surrounding the subject to provide a reliable diagnosis [4]. Because the process of auscultation in a soundproof room is not possible in most cases, especially in low-income and middle-income countries supported by a resource-poor health system, ambient denoising performed in the examination room of the health center is still a challenging task in biomedical signal processing in order to maximize the reliability of a diagnosis.

    View all citing articles on Scopus

    Styliani A. Taplidou was born in Thessaloniki, Greece in 1980. She received her Diploma degree in Electrical and Computer Engineering in 2003 from the Department of Electrical and Computer Engineering at the Aristotle University of Thessaloniki, Thessaloniki, Greece. She is currently pursuing her Ph.D. on biacoustic signals analysis at the same department. Her research interests include wavelet analysis, higher-order statistics, modeling and digital signal processing applications. She is a member of the Technical Chamber of Greece, of the IEEE and of the International Lung Sounds Association.

    Leontios J. Hadjileontiadis was born in Kastoria, Greece in 1966. He received his Diploma degree in Electrical Engineering in 1989 and the Ph.D. degree in Electrical and Computer Engineering in 1997, both from the Aristotle University of Thessaloniki, Thessaloniki, Greece.

    Since December 1999 he joined the Department of Electrical and Computer Engineering, Aristotle University of Thessaloniki, Thessaloniki, Greece as a faculty member, where he is currently an Assistant Professor, working on lung sounds, heart sounds, bowel sounds, ECG data compression, seismic data analysis, and crack detection in the Signal Processing and Biomedical Technology Unit of the Telecommunications Laboratory. His research interests are in higher-order statistics, alpha-stable distributions, higher-order zero crossings, wavelets, polyspectra, fractals, neuro-fuzzy modeling for medical, mobile and digital signal processing applications.

    Dr. Hadjileontiadis is a member of the Technical Chamber of Greece, of the IEEE, of the Higher-Order Statistics Society, of the International Lung Sounds Association, and of the American College of Chest Physicians. He was the recipient of the second Award at the Best Paper Competition of the ninth Panhellenic Medical Conference on Thorax Diseases’97, Thessaloniki. He was also an open finalist at the Student paper Competition (Whitaker Foundation) of the IEEE EMBS’97, Chicago, IL, a finalist at the Student Paper Competition (in memory of Dick Poortvliet) of the MEDICON’98, Lemesos, Cyprus, and the recipient of the Young Scientist Award of the 24th International Lung Sounds Conference’99, Marburg, Germany. In 2004 and 2005 he organized and served as a mentor to two five-student teams that have received the third and the second Award worldwide, respectively, at the Imagine Cup Competition (Microsoft), Sao Paulo, Brazil (2004)/Yokohama, Japan (2005), with the projects “SmartEyes: An enhanced navigation system for blind or visually impaired people” and “Sign2Talk: A Wearable Sign Language Translation System for Deaf or Hearing-Impaired People”, respectively.

    Dr. Hadjileontiadis also holds a Ph.D. degree in music composition (University of York, UK, 2004) and he is currently a Professor in composition at the State Conservatory of Thessaloniki, Greece.

    View full text