Respiratory sound analysis in healthy and pathological subjects: A wavelet approach

Abstract

In this paper we describe the application of a wavelet analysis-based method, to characterize the frequency power distribution of the unsteady respiratory sound signals in order to better discriminate the healthy state of a given subject. To evaluate the methodology, both normal tracheal sounds as well as adventitious respiratory sounds were investigated. In particular, our analysis shows the possibility to extract useful statistical information on the energy content and its mean frequency distribution giving us a quantitative characteristic hallmark of the respiratory pattern. The presence of sound anomalies can be pointed out through some specific patterns of the wavelet mean power spectra and thus the localization of the related quartiles which can be used as simple and efficient diagnostic indices. In this study the method has been applied in healthy subjects and patients with different respiratory diseases. Results show that different power spectra patterns characterize health from disease. Some preliminary results indicate also that pathological patterns can change as result of therapeutical interventions like mechanical ventilation.

Introduction

Complex signals analysis is a crucial issue in many diagnostic fields and the information obtained is often the key to control unwanted critical phenomena. Combustion instabilities, for example, represent the major concern in industrial power generation systems, where modern lean premixed combustors are required to operate economically and reliably with low emissions over long periods with minimal shutdown time. Acoustic emissions provide a useful diagnostic signal to characterize and to predict complex transient flames because they are strictly linked to the temporal rate of heat release change. This topic deeply stimulated many efforts to develop and to apply different advanced acoustic signal analysis, obtaining very promising results [1], [2]. The basic characteristics of these techniques suggest a wide application in signal analysis and in particular their use to complex biomedical signals appear very promising.

The total energy involved in the respiratory process comes from different sources: according to the Euler equation for a compressible fluid, we need to compare the internal work produced by muscular forces and the mechanical energy due to a ventilatory support, if any, with the sum of: (a) dissipative (frictional forces or resistive work); (b) elastic (airway compliance); (c) inertial (fluidodynamic forces); (d) acoustic (sound wave propagation). It is well known, as documented in the current wide literature, that the acoustic field gives significative information on the respiratory state, mainly due to its fluidodynamic origin, where the pressure fluctuations induced by airflow turbulence, strictly depend both on geometry and on physical properties of the involved airways [3].

Modern computer technology offers great advantages in terms of acquisition, storage and analysis of sounds that are normally heard through a stethoscope; this approach has provided new important insights into acoustic mechanisms and related quantitative diagnostics for respiratory diseases [4]. In particular, lung sound analysis to detect different kind of flow obstruction, has drawn great attention recently because the respiratory acoustic measurements have shown to be very useful in the deep investigation of different upper airways pathologies [27]. The analysis of respiratory sounds allows us to reach a considerably more information of clinical utility than that obtained by a traditional auscultation and the new findings can only be interpreted in a complete acoustical framework.

Tracheal sounds recorded at the suprasternal notch are currently the topic of significant interest because the related signal is strong with a wide range of frequencies and closely correlated to measured airflow. The main interest on the tracheal sound is due to its ability to indicate an upper airway flow obstruction and also to give a source for more or less quantitative assessments of airflow. Further, the close correlation to airflow for tracheal sound, allow us to detect some peculiar features of the more weaker lung sound at frequencies below 300 Hz where most of the acoustic lung energy resides. On the other hand, an high frequency anomalous amplitude detected on the tracheal spectrum, is a clear indication of a pathological state for the whole respiratory system [4]. The generation of tracheal sound is mainly due to turbulent air flow in upper airways which causes pressure fluctuations and sound waves within the fluid. Because of the relatively short distance between the sensor and the different sources in the upper airways, tracheal sound is often considered the less filtered breath sound.

Tracheal sound have been characterized as broad spectrum noise covering a frequency range from near 100 Hz to more than 1500 Hz with a sharp drop in power above a frequency around 800 Hz. However, the spectral shape of tracheal sound is highly variable between various subjects and thus its parametric representation is more complex than that of other breath sounds. For this reason it is necessary to develop suitable mathematical and numerical methodologies in order to extract valuable information about respiratory health.

Respiratory sounds are highly non-stationary pseudo-stochastic, or non-linear, signals due to variations of airflow rate and airflow volumes during respiration cycle. The non-linear character of the respiratory sound is mainly due to the complex turbulent flow dynamics and its structural interaction with the larger airway walls. For that reason wavelets provide a good method of decomposing these signals both in scale and time. During the last decade, they have been applied in different areas of mathematics, engineering, biology and medicine [5], [6]. Because of their suitability for analyzing transient and non-stationary signals, they have become a powerful tool as an extension to the Fourier method, in different real applications mainly for helping in the recognition and detection of typical diagnostic features. Most biomedical signals are not stationary with highly complex time–frequency characteristics. Generally, they consist of short high-frequency components closely spaced in time, followed by long-term low-frequency components closely spaced in frequency. The wavelet method can provide both very good time resolution at high frequency and good scale resolution at low frequency. This property, together with the redundancy of information inherent in continuous wavelet signal representation, makes wavelets a powerful tool in medical research and diagnostic, like studies of precursors of heart disease [7], [8], [9], studies of brain response to evoked potentials, mainly for an early detection of Alzheimer’s disease [10], [11], studies in human visual channels [12], [13], investigations for singularity detection in pulmonary microvascular pressure transients [14] and analysis of surface electromyographic signals, for Parkinson’s disease analysis and diagnostics [15], [16] and for respiratory sound analysis [17], [18], [19], [20], [21]. The main aim of the present work is to give a further contribution on how the wavelet methodology can be efficiently used to study and to characterize complex multiscale behavior of respiratory sound signals.

There is a need of non-invasive monitoring in respiratory physiology. Monitoring of respiratory parameters is essential in the treatment of acute and chronic respiratory diseases. In particular, evaluation of the work of breathing during spontaneous and assisted ventilation is an essential aspect of monitoring of effectiveness of mechanical ventilation. Rather interestingly, adequate monitoring is available in intubated mechanically ventilated patients but is rarely performed in non-intubated spontaneously breathing patients or in patients undergoing non-invasive mechanical ventilation (NIMV). Many tools for non-invasive assessment of respiratory activity have been proposed among the Respiratory Inductance Plethysmograph, Magnetometers, Optical Methods, Electromyography [22]. Unfortunately many of the measurements of breathing pattern and work of breathing in the acute patient are not accurate. Furthermore at the present there is no way to reliably and non-invasively monitor the work of breathing. At the present, many patients undergo non-invasive mechanical ventilation without an acceptable level of monitoring. Some ventilators have the facilities of monitoring tidal volume, pressures, flows, etc. However, such devices have currently no facility to monitor the degree of the respiratory muscle unloading, induced by non-invasive mechanical ventilation. One of the problems that currently limits a wide application of the non-invasive respiratory sound analysis is the technical difficulty of capturing a reliable signal from the surface of the body. Indeed no sensor is ideal and many technical problems related to a good sound capture remain to be solved.

In this work, we describe how a specific use of the wavelet analysis on broad-spectrum tracheal sound signals, it is able to distinguish between healthy subjects and patients with different respiratory diseases. More precisely, as a first step, we investigate here the ability of our wavelet-based method to quantitatively describe the respiratory state using a very synthetic vector index (the wavelet quartiles). In particular, the “normalization” effect of a non-invasive ventilatory support, on a given patient, can be now precisely quantified. Further work, also using the cluster analysis, will be devoted to see if and how this method can usefully distinguish between different respiratory pathologies. For this next step of our analysis we need a wider statistical sample of different patients, in order to point out specific features on the acoustic energy distribution in the related wavelet spectra.