COMPUTER AIDED LUNG SOUND ANALYSIS

A Preliminary Study to Assess its Potential as a

New Outcome Measure for Respiratory Therapy

Alda Marques

1, 2

, Anne Bruton

and Anna Barney

Escola Superior de Saúde da Universidade de Aveiro, Universidade de Aveiro, 3810-193 Aveiro, Portugal

School of Health Sciences, University of Southampton, Southampton, SO17 1BJ, U.K.

Institute of Sound and Vibration Research, University of Southampton, Southampton, SO17 1BJ, U.K.

Keywords: Physiotherapy, Outcome measures, Lung sounds, Crackles.

Abstract: A barrier to assess the relative effectiveness of respiratory therapies has been insufficient accurate, reliable,

and sensitive outcome measures. Lung sounds provide useful information for assessing and monitoring

respiratory patients. However, standard auscultation is too subjective to allow them to be used as an

outcome measure. In this paper, Computer Aided Lung Sound Analysis (CALSA) characterising crackles’

Initial Deflection Width (IDW) and Two Cycle Deflection (2CD) is proposed as a potential objective, non-

invasive, bedside outcome measure to assess the response to alveolar recruitment and airway clearance

interventions. A preliminary ‘repeated measures’ experimental study was conducted. Seventeen participants

with cystic fibrosis were recruited from out-patient clinics. Demographic, anthropometric and lung sound

data were collected. The intra-subject reliability of crackles’ IDW and 2CD was found to be ‘good’ to

‘excellent’, estimated by the Analysis of Variance, Intraclass Correlation Coefficient, Bland and Altman

95% limits of agreement and Smallest Real Difference. It is concluded that crackle IDW and 2CD detected

by CALSA are reliable and stable measures. In future, CALSA may be useful for assessing and monitoring

respiratory interventions in clinical settings.

1 INTRODUCTION

The lack of reliable outcome measure has been an

obstacle to providing evidence about the relative

effectiveness of respiratory therapy interventions.

The main aims of respiratory physiotherapy include:

increasing alveolar recruitment, thereby improving

ventilation; increasing secretion removal and

therefore airway clearance; decreasing work of

breathing and consequently dyspnoea; increasing

muscle strength and endurance to increase exercise

capacity and independence in daily functioning and

increasing patients’ understanding of their lung

condition to promote self-management. Research

into airway clearance techniques has been one of the

priorities identified for research (CSP, 2002). In this

paper we explore the potential for computer aided

lung sound analysis (CALSA) to be used as an

outcome measure for respiratory therapy

interventions.

Respiratory physiotherapists currently use the

following outcome measures to monitor their

interventions and evaluate their practice: sputum

quantity, respiratory function tests, tests of gas

exchange, imaging evidence and standard

auscultation techniques. Most of these are not

specifically related to the physiotherapy intervention

employed and may be affected by other factors

(Marques et al., 2006). There is no gold standard

outcome measure that is specifically related to

respiratory physiotherapy interventions. Most of the

published respiratory physiotherapy research

compares two or more active interventions rather

than an active intervention versus an inactive

control. In such studies it is never clear if differences

are not detected because the outcome measures are

not appropriate, or because the treatments being

compared are equally effective/ ineffective.

Although there are other more invasive or

laboratory-based outcome measures available, these

are generally only applicable to a research setting.

251

Marques A., Bruton A. and Barney A. (2009).

COMPUTER AIDED LUNG SOUND ANALYSIS - A Preliminary Study to Assess its Potential as a New Outcome Measure for Respiratory Therapy .

In Proceedings of the International Conference on Health Informatics, pages 251-256

DOI: 10.5220/0001547002510256

 SciTePress

1.1 Lung Sounds

Normal lungs generate breath sounds as a result of

turbulent airflow in the trachea and proximal

bronchi. The airflow in the small airways and alveoli

has a very low velocity and is laminar, and therefore

silent. Turbulent flow characteristics are influenced

by airway dimensions, which are a function of body

height, body size, age, gender and airflow will all

affect breath sounds (Pasterkamp et al., 1997).

Sounds heard or recorded at the chest differ

according to the location at which they are heard or

recorded, and vary with the respiratory cycle

(Sovijarvi et al., 2000). The geometry of the bronchi

also contributes to the complexity of the thoracic

acoustics (Kompis et al., 2001) because it affects

flow, and consequently breath sounds.

Normal breath sounds can be classified as

‘abnormal’ if heard at inappropriate locations. There

are also added sounds (known as adventitious

sounds) which can be continuous (wheezes) or

discontinuous (crackles). Their presence usually

indicates a pulmonary disorder.

Wheezes are continuous adventitious lung

sounds. The mechanisms underlying their

production appear to involve an interaction between

the airway wall and the gas moving through the

airway (Meslier et al., 1995), producing continuous

undulating sinusoidal deflections by fluttering of the

airway walls. Wheezes are clinically defined as

musical sounds and can be characterised by their

location, intensity, pitch, duration in the respiratory

cycle, and relationship to the phase of respiration

(Meslier et al., 1995).

Crackles are discontinuous adventitious sounds

and their presence may be an early sign of

respiratory disease. They are intermittent, non-

musical, brief sounds thought to be caused by the

acoustic energy generated by pressure equalization

or change in elastic stress after a sudden opening or

closing of airways (Forgacs, 1978, Nath and Capel,

1974). Their character is explosive and transient and

depends on the diameter of the airways. Their short

duration and often low intensity, makes their

discrimination and characterisation by normal

auscultation very difficult (Kiyokawa et al., 2001).

Crackles are generally characterised by the Initial

Deflection Width (IDW), i.e. the duration of the first

deflection of the crackle and the Two Cycle

Deflection (2CD), i.e. the duration of the first two

cycles of the crackle. The mean values of IDW and

2CD durations for fine and coarse crackles are

defined (ATS, 1977, Sovijarvi et al., 2000). During

respiratory disease, the involvement of different

airways is associated with the crackle frequency, i.e.,

high frequency crackles are associated with

peripheral airways and lower frequency crackles

with upper airways (Fredberg and Holford, 1983).

These measurements therefore have the potential to

be useful as an outcome measure for respiratory

therapy.

1.2 Standard Auscultation

Standard auscultation via a stethoscope is an

assessment tool used by many health professionals

during chest examination in their clinical practice,

and is often used by physiotherapists to monitor

patients’ response to respiratory interventions.

However, the literature has contradictory reports

about its value in routine current practice. Some

authors argue that auscultation is an inappropriate

outcome measure because of the differences in

health professionals’ hearing acuity as well as in the

properties of stethoscopes. There can also be

different approaches to the description of

auscultatory findings, nomenclature difficulties, and

inter- and intra-observer variability (Welsby et al.,

2003) Others have argued that auscultation is an

easy, rapid, effective, non-invasive, and cost-

effective way of assessing the condition of the

airway and breathing (Chen et al., 1998). However,

agreement between observers during standard

stethoscope examination for the presence of normal

or adventitious lung sounds was found to be only

‘poor to moderate’, and clinical experience was not

found to have any clear effect on accuracy or

reliability (Brooks et al., 1993). Elphick et al. (2004)

found that using computerised acoustic analysis of

recorded lung sounds improved the reliability of

detection for all sounds when compared to listening

through a stethoscope. Therefore, although the use

of a standard stethoscope may be too subjective to

provide a useful outcome measure, the sounds

generated from the lungs provide useful information,

and relate directly to movement of air and

secretions.

1.3 Computer Aided Lung Sound

Analysis

There is a great deal of information derivable from

lung sounds, that is not normally readily accessible

even to experienced clinicians and exceeds the

memory capacity of most people. Lung sounds

interpretation is enhanced using CALSA through the

efficient objective data collection, generation of

permanent records of the measurements made with

HEALTHINF 2009 - International Conference on Health Informatics

252

easy retrievability and through graphical

representations that help with diagnosis and

management of patients’ suffering from chest

diseases (Earis and Cheetham, 2000). Digital

recordings of normal lung sounds have shown high

inter- and intra-subject repeatability with the inter-

individual variability explained by height, gender

and anatomic characteristics (Sanchez and Vizcaya,

2003).

There is some evidence in the literature to

support the hypothesis that CALSA characterising

adventitious lung sounds may be a useful outcome

measure. The number and distribution (per breath)

of adventitious sounds has been associated with

severity of disease and crackles characteristics have

been found to differ in different diseases, allowing

differentiation between conditions (Piirila, 1992).

Furthermore, the number of wheezes per respiratory

cycle has been reported as a good indicator for

airway obstruction (Baughman and Loudon, 1984).

CALSA has already been used to assess the airways’

response to bronchodilators and bronchoconstrictors

in children and in adults. Malmberg et al. (1994)

studied 11 asthmatic children (aged 10 to 14 years)

and found that spectral analysis of lung sounds can

be used to detect airways obstruction during

bronchial challenge tests. When combined with

spirometry, CALSA increased the sensitivity of

detection of pulmonary disease and was able to

provide early signs of lung disease that was not

detected by spirometry alone (Gavriely et al., 1994).

The use of CALSA has been found to be

specific, reliable, and sensitive within the limited use

to which it has been put to date but has not yet been

evaluated as an outcome measure for physiotherapy.

Although it is known that normal lung sounds are

reliable and that adventitious lung sounds have some

clinical meaning, the reliability of the specific

parameters of adventitious lung sounds (e.g.

crackles’ IDW and 2CD) has not been adequately

explored. In order for CALSA to be used as a valid

outcome measure its reliability amongst patient

populations in clinical settings must be

demonstrated. Therefore, the inter- and intra-subject

reliability of CALSA to measure crackles over short

time-periods in adult patients with cystic fibrosis

were explored.

2 METHODOLOGY

A single group repeated measures design was

employed in which triplicate recordings were made

from the same patients over a short time period. The

study received full approval from Southampton &

South West Hampshire Research Ethics Committees

(A).

Since this was a reliability study, power

calculations were not appropriate. Potential

participants were identified via the respiratory out-

patient clinics held at two hospitals on the south

coast of the UK. Participants were included in the

study if they were 1) able to give and sign informed

consent; 2) formally diagnosed with cystic fibrosis

(CF); 3) 18 years of age or older and 4) clinically

stable for one month prior to the study (no hospital

admissions, exacerbations/ infections or change in

medication). Participants were excluded from the

study if they had co-existing lung pathologies.

Anthropometric data such as height and weight

were measured using calibrated digital scales. The

lung sound recordings were performed with a digital

stethoscope (WelchAllyn Meditron, 5079-402). The

input from the microphone was connected via an

amplifier to a laptop with customised software,

suitable for data acquisition, written in Matlab

(version 7.1).

2.1 Protocol

All participants provided informed consent prior to

data collection. Demographic and basic

anthropometric data were recorded first. The lung

sound recordings followed the guidelines defined by

Computerized Respiratory Sound Analysis for short

term acquisition (Sovijarvi et al., 2000).

Participants’ skin was marked with a pen in seven

different places to ensure consistency of stethoscope

placement: one centrally over the trachea (sternal

notch); two on the back of the chest (right and left

bases: at 5 cm from the paravertebral line and 7 cm

below the scapular angle); two on the front of the

chest (right and left: in the second intercostal space,

mid-clavicular line); two on the side of the chest

(right and left: in the fourth to fifth intercostal space,

mid axillary line). Lung sound recordings were then

performed with the digital stethoscope held by hand

over each location. Participants were asked to

breathe through the mouth during recordings. Three

sets of recordings were made for 25 seconds at each

marked location.

2.2 Analysis

Gender, date of birth, height and weight data were

entered into SPSS version 14. Body mass index

(BMI) in (kg/m

) was calculated using the formula

COMPUTER AIDED LUNG SOUND ANALYSIS - A Preliminary Study to Assess its Potential as a New Outcome

Measure for Respiratory Therapy

253

BMI = weight/(height)

. Descriptive statistics were

used to describe the sample.

The lung sounds were recorded with a sampling

frequency of 44.1 KHz. All files from the seven

anatomical sites were processed to detect crackles

using algorithms written in Matlab based on

Vannuccini et al.’s algorithm (1998). The data from

all files with the duration in milliseconds of the

crackles variables (IDW and 2CD) were imported to

SPSS version 14 for statistical analysis.

2.2.1 Relative Reliability

Inter-subject reliability was analysed using analysis

of variance (ANOVA). All recordings were

performed by the same researcher. The inter- and

intra-subject reliability was examined in each

recording position in each session. Therefore, the

ICC was calculated using the equation (1,k) which

uses the one-way ANOVA table: ICC (1,K) =

(BMS-WMS)/BMS, where ICC is the Intraclass

Correlation Coefficient, BMS = between subjects

mean squares; WMS = within subjects mean

squares; k = number of observers or measures. The

ICC results were analysed according to Fleiss (1986)

criteria i.e. values above 0.75 represent excellent

reliability; between 0.4 and 0.75 represent moderate

to good reliability and below 0.4 represent poor

reliability.

2.2.2 Absolute Reliability

Bland-Altman plots were performed to provide

visual information about systematic bias and random

error by examining the direction and magnitude of

the scatter around the mean difference (Bland and

Altman, 1986). The standard error of measurement

(SEM) was obtained by calculating the square root

of the within subject mean square (WMS) values

obtained in the ANOVA table performed for each

recording position for each session. These values

were then used to calculate the Smallest Real

Difference (SRD), which represents the smallest

change that can be interpreted as a real difference.

3 RESULTS

Seventeen CF participants, age range of 18 to 67

years old (8 female and 9 male) and average BMI of

20.6±3.3 kg/m

, were recruited. The inter-subject

reliability analysis confirmed the expected

variability between subjects. It has been suggested

that to assess intra-rater (or test-retest) or inter-rater

reliability in reliability studies, a measure of relative

reliability, the ICC, and a measure of absolute

reliability, Bland & Altman 95% limits of

agreement, should both be reported (Rankin and

Stokes, 1998).

The ICC values range between 0.57 to 0.85 for

the crackles’ IDW variable and between 0.75 and

0.96 for the crackles’ 2CD. The ICC results were

generally found to be excellent in both groups of

participants. The crackles’ IDW variable generally

presented a lower ICC value when compared with

the crackles’ 2CD variable. Bland and Altman 95%

limits of agreement were performed and the scatter

plots were analysed in all recording positions for

both groups of participants on the three different

occasions. No systematic bias was present. The SRD

ranged from 0.33 to 0.76 ms for the crackles’ IDW

and from 1.29 to 2.66 ms for the crackles’ 2CD.

4 DISCUSSION

The inter- and intra-subject reliability of crackles’

IDW and 2CD measured using CALSA has been

rarely investigated. There is, however, a consensus

that the inter and intra-observer reliability of the

detection of added lung sounds among health

professionals using tape-recorders, audio-files or

standard auscultation is generally poor (Elphick et

al., 2004). The inter-subject variability for both

crackles’ variables studied (IDW and 2CD) in this

research was found to be high, as shown by the

significant ANOVA. This high inter-subject

variability for crackles’ IDW and 2CD duration was

expected due to differences in demographic and

anthropometric characteristics (Sanchez and

Vizcaya, 2003), and varying acuity of pathology.

The intra-subject reliability was found to be

‘good’ to ‘excellent’ with no systematic bias

between the repeated measures. Previous studies

have also reported high intra-subject reliability of

lung sounds in healthy subjects (Mahagnah and

Gavriely, 1994) and in healthy subjects and patients

with fibrosing alveolitis (Sovijarvi et al., 1996).

However, these studies have analysed lung sounds in

the frequency domain, in small samples of mainly

healthy subjects and have not included people with

excessive secretions making comparisons difficult.

The ICC and Bland and Altman 95% limits of

agreement have been recommended as the more

adequate methods to assess reliability (Rankin and

Stokes, 1998). The ICC for the crackles’ IDW and

2CD suggested ‘excellent’ or ‘good’ reliability in

almost all recording positions, in both groups of

HEALTHINF 2009 - International Conference on Health Informatics

254

participants. However, the ICC should be interpreted

with caution. Bland and Altman 95% limits of

agreement are independent of the true variability in

the observations and therefore, complement the ICC

analysis and provide detail regarding the nature of

the observed intra-subject variability (Rankin and

Stokes, 1998). The reliability assessed from Bland

and Altman techniques was found to be acceptable,

and no consistent bias was detected in any recording

position of the two crackles’ variables studied.

Therefore, crackle characterisation using CALSA

was deemed reproducible over short time periods.

The ICC has rarely been used when analysis lung

sounds, but ‘satisfactory’ reproducibility of lung

sounds based on ICC results have been reported

(Schreur et al., 1994). However, this analysis was

performed in the frequency domain where spectral

characteristics were considered and the recordings

were performed in a sound proofed room.

In this research, the SRD values, over short time

periods, for both variables studied (crackles’ IDW

and 2CD) presented a similar range of values

indicating the stability of the measure in CF

participants. Smallest real difference calculations

were not found in other published studies involving

lung sound analysis and therefore, comparisons with

the findings of this research were not possible.

As it has been demonstrated, the use of CALSA

in terms of clinical properties shows potential to be

use as an outcome measure for respiratory therapy

but this has not been previously explored. At this

point in time, the data related to lung sounds are

complex and time consuming to analyse. To be

clinically useful as an outcome measure it will be

essential to simplify, and increase the speed of, the

analytical process. Furthermore the cost-

effectiveness of implementing this outcome measure

is unknown. Validation of CALSA as a responsive

outcome measure is challenging because of the lack

of a gold standard respiratory therapy measure with

which to compare it. Studies relating to

responsiveness to change of lung sounds before and

after an intervention of known effect e.g.

bronchodilators, suction, would help in clarifying

the responsiveness of the measure, and increasing

understanding of the validity of CALSA in clinical

settings. Nevertheless, the aims proposed at the

outset for this research have been achieved.

5 CONCLUSIONS

This preliminary study suggests that the use of

CALSA to characterise crackles is reproducible over

short time periods in cystic fibrosis outpatients. This

finding gives initial confidence that CALSA has

potential as an outcome measure for respiratory

therapy interventions (such as physiotherapy for

airway clearance) but further evaluation is

necessary.

ACKNOWLEDGEMENTS

We would like to thank the Fundação para a Ciência

e Tecnologia (FCT), Portugal, for funding one of the

authors (Alda Marques) during her PhD studies.

REFERENCES

A.T.S. (1977) Update nomenclature for membership

relation. ATS News, 3, 5-6.

Baughman, R. P. & Loudon, R. G. (1984) Quantitation of

wheezing in acute asthma. Chest, 86, 718-722.

Bland, J. M. & Altman, D. G. (1986) Statistical methods

for assessing agreement between two methods of

clinical measurement. Lancet, 1, 307-310.

Brooks, D., Wilson, L. & Kelsey, C. (1993) Accuracy and

reliability therapists in auscultating tape-recorded lung

sounds. Physiother Canada, 45, 21-24.

Chen, S.-C., Chang, K.-J. & Hsu, C.-Y. (1998) Accuracy

of auscultation in the detection of

haemopneumothorax. European Journal Surgery, 164,

643-645.

CSP (2002) Priorities for Physiotherapy Research in the

UK: Topics prioritised by the cardiorespiratory expert

panel (Annex 1), London, Chartered Society of

Physiotherapy.

Earis, J. E. & Cheetham, B. M. G. (2000) Current methods

used for computerized respiratory sound analysis.

European Respiratory Review, 10, 586-590.

Elphick, H. E., Lancaster, G. A., Solis, A., Majumdar, A.,

Gupta, R. & Smyth, R. L. (2004) Validity and

reliability of acoustic analysis of respiratory sounds in

infants. Archives Disease Childhood, 89, 1059-1063.

Fleiss, J. (1986) Reliability of measurements, New York,

John Wiley & Sons.

Forgacs, P. (1978) Lung sounds, London, Bailliere

Tindall.

Fredberg, J. & Holford, S. K. (1983) Discrite lung sounds:

crackles (rales) as stress-relaxation quadrupoles.

Journal of Acoustic Society of America, 73, 1036-

1046.

Gavriely, N., Nissan, M., Cugell, D. W. & Rubin, A. H. E.

(1994) Respiratory health screening using pulmonary

function tests and lung sound analysis. European

Respiratory Journal, 7, 35-42.

Kiyokawa, H., Geenberg, M., Shirota, K. & Pasterkamp,

H. (2001) Auditory detection of simulated crackles in

breath sounds. Chest, 119, 1886-1892.

COMPUTER AIDED LUNG SOUND ANALYSIS - A Preliminary Study to Assess its Potential as a New Outcome

Measure for Respiratory Therapy

255

Kompis, M., Pasterkamp, H. & Wodicka, G. R. (2001)

Acoustic imaging of the human chest. Chest, 120,

1309-1321.

Mahagnah, M. & Gavriely, N. (1994) Repeatability of

measurements of normal lung sounds. American

Journal Respiratory Critical Care Medicine, 149, 477-

481.

Malmberg, L. P., Sovijarvi, A. R. A., Paajanen, E., Piirila,

P., Haahtela, T. & Katila, T. (1994) Changes in

frequency spectra of breath souds during histaminic

challenge test in adult asthmatics and healthy control

subjects. Chest, 105, 122-132.

Marques, A., Bruton, A. & Barney, A. (2006) Clinically

useful outcome measures for physiotherapy airway

clearance techniques: a review. Physical Therapy

Reviews, 11, 299-307.

Meslier, N., Charbonneau, G. & Racineux, J.-L. (1995)

Wheezes. European Respiratory Journal, 8, 1942-

1948.

Nath, A. R. & Capel, L. H. (1974) Inspiratory crackles and

mechanical events of breathing. Thorax, 29, 695-698.

Pasterkamp, H., Kraman, S. S. & Wodicka, G. R. (1997)

Respiratory sounds – advances beyond the

stethoscope. American Journal Respiratory Critical

Care Medicine, 156, 974-987.

Piirila, P. (1992) Changes in crackle characteristics during

the clinical course of pneumonia. Chest, 102, 176-183.

Rankin, G. & Stokes, M. (1998) Reliability of assessment

tools in rehabilitation: an illustration of appropriate

statistical analyses. Clinical Rehabilitation, 12, 187-

199.

Sanchez, I. & Vizcaya, C. (2003) Tracheal and lung

sounds repeatability in normal adults. Respiratory

Medicine, 97, 1257-1260.

Schreur, H. J. W., Vanderschoot, J., Zwinderman, A. H.,

Dijkman, J. H. & Sterk, P. J. (1994) Abnormal lung

sounds in patients with asthma during episodes with

normal lung function. Chest, 106, 91-99.

Sovijarvi, A. R. A., Malmberg, L. P., Charbonneau, G.,

Vanderschoot, J., Dalmasso, F., Sacco, C., Rossi, M.

& Earis, J. E. (2000) Characteristics of breath sounds

and adventitious respiratory sounds.

European

Respiratory Review, 10, 591-596.

Sovijarvi, A. R. A., Malmberg, P., Paajanen, E., Piirila, P.,

Kallio, K. & Katila, T. (1996) Averaged and timed-

gated spectral analysis of respiratory sounds -

Repeatability of spectral parameters in healthy men

and in patients with fibrosing alveolitis. Chest, 109,

1283-1290.

Vannuccini, L., Rossi, M. & Pasquali, G. (1998) A new

method to detect crackles in respiratory sounds.

Technology and Health Care, 6, 75-79.

Welsby, P. D., Parry, G. & Smith, D. (2003) The

stethoscope: some preliminary investigations.

Postgraduate Medical Journal, 79, 695-698.

HEALTHINF 2009 - International Conference on Health Informatics

256