(Chest. 2001;119:1886-1892.)
© 2001
American College of Chest Physicians
Auditory Detection of Simulated Crackles in Breath Sounds*
Hiroshi Kiyokawa, MD;
Matthew Greenberg;
Kazuhiko Shirota, MD and
Hans Pasterkamp, MD
*
From the Department of Pediatrics and Child Health, University of Manitoba, and the Respiratory Acoustics Laboratory, John Buhler Research Centre, Winnipeg, Canada.
Correspondence to: Hiroshi Kiyokawa, Room 510, John Buhler Research Centre, 715 McDermot Ave, Winnipeg, Manitoba R3C 3J7, Canada; e-mail: kiyokawa{at}cc.umanitoba.ca
 |
Abstract
|
|---|
Background: Computerized analysis of breath sounds has
relied on human auditory perception as the reference standard for
identifying crackles. In this study, we tested the human audibility of
crackles by superimposing artificial clicks on recorded breath sounds
and having physicians listen to the recordings to see if they could
identify the crackles.
Objectives: To establish the
audibility of simulated crackles introduced in breath sounds of
different intensity, to study the effects of crackle characteristics on
their audibility, and to investigate crackle detection within and
between observers.
Methods: Fine, medium, and coarse
crackles with large and small amplitude were synthesized by computer
software. Waveform parameters were based on published characteristics
of lung sound crackles. The amplitude for small crackles was defined as
just above the threshold of audibility for simulated crackles inserted
in sound recorded during breath hold. Simulated crackles were then
superimposed on breath sounds recorded at 0 L/s (breath hold), 1 L/s,
and 2 L/s airflow. Five physicians listened during playback on two
separate occasions to determine if crackles could be heard and to
calculate the interobserver and intraobserver variations.
Results: Failed detection of crackles was significantly
more common in the following conditions: (1) background breath sounds
had higher intensity (2 L/s airflow) compared to lower intensity (1
L/s), (2) crackle type was coarse or medium compared to fine, and (3)
crackle amplitude was small compared to large. Both intraobserver and
interobserver agreements were high (
> 0.6).
Relevance: The validation of automated techniques for
crackle detection in lung sound analysis should not rely on
auscultation as the only reference. Detection of crackles is
facilitated when patients take slow, deep breaths that generate little
breath sounds.
Key Words: auditory perception • auscultation • computer analysis • crackles • respiratory sounds
|
Introduction
|
|---|
Crackles
(rales) are useful indicators of cardiorespiratory disease. The timing,
pitch, and waveform of crackles reflect different pathophysiology in
diseases,1
such as pneumonia, bronchiectasis, asbestosis,
sarcoidosis, fibrosing alveolitis, cystic fibrosis, and pulmonary
congestion due to cardiac failure.2
3
4
5
6
Since 1989, there
have been several attempts7
8
9
10
11
12
13
14
15
16
to detect crackles by
automated methods. Some of the investigators employed human
auditory identification as the reference standard in developing
computerized lung sound analyzers. Considering the known limitations of
the human auditory system to identify clicks under different
circumstances,17
18
19
we questioned if auscultation is a
reliable reference standard for this identification and hypothesized
that crackles may be masked by normal breath sounds. In this study, we
introduced various simulated crackles into normal breath sounds that
had been recorded at different airflows. The resulting audio files with
defined timing and characteristics of crackles became our reference
standard. When playing these audio files to different observers, we
investigated the audibility of crackles in breath sounds of varying
loudness, the effect of crackle type and amplitude on their audibility,
and the agreement of crackle detection within and between different
observers.
|
Materials and Methods
|
|---|
Generation of Simulated Crackles
Crackles were generated by mathematical functions (MATLAB;
MathWorks; Natick, MA) designed to provide waveforms (Fig 1
, Appendix) that were similar to those reported for naturally occurring
crackles.20
21
22
Medium crackles were designed to fall
midway between the reported characteristics of fine and coarse
crackles. To decide on the amplitude for small fine crackles, we
performed a preliminary study: crackles with various amplitudes were
generated and superimposed on chest sounds recorded at zero airflow
(breath hold). Two of us (H.K., H.P.) listened and determined
the smallest amplitude that could be detected consistently. Large
crackles were then defined to have twice the amplitude of small
crackles.
View larger version (15K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1.. Simulated crackles. Simulated crackles had the
following characteristics: Top, a: fine (IDW = 0.5 ms;
two-cycle duration [2CD] = 5.0 ms). Middle, b:
medium (IDW = 0.9 ms; 2CD = 6.0 ms). Bottom, c:
coarse (IDW = 1.2 ms; 2CD = 9.0 ms).
|
|
Respiratory Sound Recording
Sounds during breath hold (zero flow) and at target flows of 1
L/s and 2 L/s were recorded with a computerized system for recording
and analysis (Table 1
). Recording was performed in a sound-insulated chamber. An electret
microphone was attached with a coupling chamber23
on the
left lower back of a healthy male volunteer, using double-sided
adhesive tape. Airflow at the mouth was measured with a
pneumotachometer. The chest sounds were recorded during breath hold and
during breathing at target airflows of 1 L/s and 2 L/s ± 20%
tolerance. Breath sounds were amplified and filtered, and the signals
were digitized and stored on computer. The loudest respiratory sound
level (at 2 L/s) was 26% of the 12-bit linear quantization range. From
this recording, consecutive segments of 12 s (breath hold) or six
breaths (1 L/s and 2 L/s airflow) without audible clicks or crackles
were carefully selected. These parts were then saved as sound files of
different amplitude (breath hold and breath sounds at 1 L/s and 2 L/s).
Frequency spectra of these sounds are shown in Figure 2
. At 500 Hz, lung sound intensity was approximately 15 decibels (dB) and
30 dB above background noise (breath hold reference) at 1 L/s and 2 L/s
airflow, respectively.
View larger version (24K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2.. Breath sounds. Frequency spectra of
(a) breath hold, (b) 1 L/s airflow, and,
(c) 2 L/s airflow. At 500 Hz, lung sound intensity was
approximately 15 dB and 30 dB at 1 L/s and 2 L/s airflow, respectively,
above the chest sounds during breath hold.
|
|
Crackles in Breath Sounds
Simulated crackles were then superimposed on the recorded sounds
(MATLAB) to generate audio files for testing (Fig 3
, Appendix). Each of 24 inspirations (six breaths x four files) of 1
L/s or 2 L/s recordings had one of simulated crackles inserted.
Crackles were fine, medium, or coarse and had a large or small
amplitude. In six breath hold recordings, crackles were introduced at
random timing and the number of crackles in each file was set to four.
Details of these recordings and the simulated crackles are shown in
Table 2 .
View larger version (25K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3.. Simulated crackle in recorded breath sounds.
Simulated coarse crackle in recorded breath sounds at
(top, a) breath hold,
(middle, b) 1 L/s airflow, and
(bottom, c) 2 L/s airflow. Thick line
indicates superimposed simulated coarse crackle. Note the relative
increase in the amplitude of sound wave deflections of normal
(vesicular) breath sounds, compared with the crackle amplitude, at
higher air flows.
|
|
Auscultation of Sound Files and Statistical Analysis
Four male physicians and one female physician listened to the 14
audio files on two separate occasions up to 3 days apart. They were 35-
to 45-year-old specialists in pediatrics and internal medicine with
normal findings on routine screening audiometry. Playback of the
acoustic files from the computer via the sound card and headphones
(Table 1)
was provided by commercial software (Cool Edit Pro;
Syntrillium Software; Scottsdale, AZ). The listening level was
set to 90% of the maximum output of the sound card, based on the
comfort level for all listeners. All tests were performed within a
sound-insulated chamber. Observers were instructed that crackles may or
may not be present in the recording. They were asked to mark on a
graph, during or immediately after listening, the approximate location
where they detected crackles. The results were analyzed and compared
with statistical software (SPSS for Windows; SPSS; Chicago, IL).
Both false-negative (observer did not detect an actual crackle) and
false-positive answers (observer heard a crackle when there was none)
were counted as "missed." values were calculated to evaluate
intraobserver and interobserver agreements.24
|
Results
|
|---|
The detection of crackles became more difficult when breath sounds
had greater intensity. The proportion of undetected crackles was
significantly higher at 2 L/s airflow (59.6%) compared to 1 L/s
(20.8%; Fig 4
). The type of crackles also had an effect on the rate of detection.
Coarse (55.0%) and medium crackles (42.2%) were more often missed
than fine crackles (17.8%). Not surprisingly, the amplitude of
crackles also affected the rate of detection. In breath sounds at 1
L/s, 41.1% of small crackles were missed. Although large crackles were
more readily identified at this airflow, still 14.4% were missed. In
breath sounds at 2 L/s airflow, the proportion of missed crackles was
very high for both small (88.9%) and large crackles (70.0%). Crackles
with small amplitude that were clearly audible during breath hold were
often undetected even at the 1 L/s airflow. The agreement between and
also within observers was strong, with values > 0.6 (Fig 5
).
View larger version (35K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4.. Crackle detection. Proportion of missed crackles
shown in relation to (left, a) airflow,
(center, b) crackle type, and
(right, c) crackle amplitude. Statistical
significances determined by  2 test.
|
|
|
Discussion
|
|---|
Human auditory perception and neuropsychological processing of
breath sounds perform well with regard to selective attention and
pattern recognition.25
However, auditory recognition can
be impaired by several masking effects. The effect of broadband noise
on the detection of clicks has been reported in studies of acoustic
sensation and of auditory brainstem responses.18
19
In the
present study, we found that the noise of normal breath sounds also has
an effect on the perception of clicks or crackles.
As expected, louder breath sounds masked crackles more effectively than
low-intensity breath sounds. Fine crackles may be more easily
recognized than coarse crackles because their waveforms differ more
clearly from those of normal breath sounds. The initial deflection
width (IDW) of fine crackles was as short as 0.5 ms, and the principal
component frequency was approximately 1 kHz, while the IDW of coarse
crackles was 1.2 ms and the frequency was approximately 400 Hz. Since
normal inspiratory breath sounds contain power mostly below 500
Hz,26
fine crackles will stand out more clearly.
We did not address differences in the masking of crackles by other
types of normal or adventitious respiratory sounds, eg,
tracheal sounds or wheezing. However, it is likely that such sounds of
greater intensity and broader range of frequencies could have an even
greater masking effect on crackles. The insertion of maximally one
crackle per breath was artificial but allowed an unambiguous evaluation
of their detection. Potentially relevant masking effects by multiple
crackles in close proximity, eg, in fibrosing lung disease
presenting with "Velcro rales," was not addressed in this study.
The simulated crackles in this study were modeled after descriptions of
crackle measurements in human subjects20
21
22
and did sound
different to the observers, depending on waveform characteristics.
However, the experimental design focused on the detection of
crackles and not on the subjective classification in categories of
fine, medium, and coarse. Thus, we cannot address the observer
agreement of crackle classification.
For the purpose of this study, we created an optimized environment
quite different from the typical clinical setting, which would not
offer sound-insulated chambers and noise-cancellation technology. The
recognition of crackles is going to be more difficult in a noisy
setting, eg, in a busy emergency department. Furthermore,
the observers in our study could anticipate the presence of crackles
and focus exclusively on their detection. Some crackles were
practically inaudible when breath sounds were present. In contrast
to the normal clinical situation, the presence of these crackles was
known and this could be taken into account when calculating the
values. All these factors explain the good observer agreement that was
higher than previously reported for the auscultatory detection of
crackles.27
28
29
Only Workum et al30
found
values > 0.6 in a study of adventitious sound recognition in lung
sound recordings, and they also used a standardized and optimized
environment.
Interestingly, we found that crackles of the same type and amplitude,
inserted in lung sound recordings of the same airflow, could be either
audible or inaudible, with complete agreement between observers (Fig 6 ). The audibility was apparently determined by the waveform
characteristics of the surrounding normal breath sounds (Fig 7
). At present it is impossible to know whether this situation also
occurs in vivo, ie, the merging of crackle sound
waves with breath sounds in a way that would completely mask their
presence. Techniques for detection other than subjective auscultation
are required to resolve this issue. Various methods for automated
crackle detection have shown promise.7
8
9
10
11
12
13
14
15
16
Crackle
quantification by computerized lung sound analysis is typically
compared to subjective auscultation as the reference
standard.10
14
16
It will be a challenge to devise testing
conditions for their validation when auscultation cannot be relied on.
Visual identification of characteristic crackle waveforms in time
expanded displays of lung sound recordings also has to be compared
under controlled circumstances, ie, where the presence and
timing of crackles are known a priori.
View larger version (21K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6.. Detectability of crackles. Each one of 72 crackles
used in this study was assessed 10 times (five observers x two
assessments). We stratified the crackles by rate of detection.
Accordingly, crackles can be classified into two types: detectable and
undetectable. Detectable crackles were recognized almost all the time
(at least 9 of 10 assessments by five observers), whereas undetectable
crackles were hardly recognized at all (in 1 of 10 assessments by five
observers or less).
|
|
View larger version (26K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 7.. Detected and undetected crackle.
Top, a: simulated coarse, large amplitude
crackle that had been consistently detected by all observers.
Bottom, b: same simulated crackle,
superimposed at a different point of the same lung sound recording (2
L/s), but never detected by any observer.
|
|
Technological solutions may be found to improve the detection of
crackles in breath sounds. This may lead to new diagnostic methods,
considering that the inspection of lung sound waveforms to detect
crackles can be more sensitive than chest radiographs to detect signs
of early asbestosis.31
While this technology is being
developed, physicians will continue to use their stethoscopes to detect
crackles. They should be aware of the masking effect of normal breath
sounds if the focus is on the detection of crackles. Since the
generation of crackles depends more on lung volume changes than on
airflow,32
33
patients should be advised to take slow and
deep breaths in order to minimize flow turbulence and thus reduce the
intensity of normal breath sounds. Thus, the application of insights
gained from respiratory acoustic measurements and a better
understanding of the psychoacoustics of auscultation may improve the
reliability and therefore the value of chest physical examination.
|
Appendix 1
|
|---|
Crackle Simulation
To define a waveform y(t), 0 
t 1, which has the
properties associated with a crackle, y(t) was defined so that:
1. the crackle has two cycles.
2. the location of the first positive t-intercept of y(t) appears
explicitly as a parameter in the formula.
3. most of the power in y(t) is concentrated near the beginning of
the waveform.
4. the resulting waveform looks authentic to experienced
observers.
A simple function possessing two cycles and its first
positive t-intercept at t0 is given by
y0(t) = sin(4
t
),
where = log(0.25)/log(t0).
To shift most of the power to the beginning of the
waveform, a modulating function m(t) = 0.5{1 + cos
[2(t0.5 - 0.5)]} was applied. The
waveform defined by y(t) = m(t)y0(t) satisfies
condition 4 (above) in that it looks authentic to experienced
observers.
To Insert a Simulated Crackle
To insert a crackle wave into an inspiration bounded
between t0 and t3,
the respiratory phase was divided into thirds, defining
t1 = t0 + (t3 - t0)/3
and
t2 = t0 + 2(t3 - t0)/3.
The superposition began at
t1 + r(t2 - t1)
where r is a random number between 0 and 1. This ensured that the
crackle wave was inserted into the noisiest part of inspiration, close
to the peak tidal flow rate.
|
Footnotes
|
|---|
Abbreviations: dB = decibel;
IDW = initial deflection width
Financial support provided by the Manitoba Lung Association
(Fellowship awards, Drs. Shirota and Kiyokawa), and the
Children’s Hospital Foundation of Manitoba (Summer Studentship, Mr.
Greenberg).
Presented at the 96th American Thoracic Society International
Conference, Toronto, Canada, May 7, 2000.
Received for publication July 18, 2000.
Accepted for publication November 29, 2000.
|
References
|
|---|
-
Piirila, P, Sovijarvi, AR (1995) Crackles: recording, analysis and clinical significance. Eur Respir J 8,2139-2148[Abstract]
-
Nath, AR, Capel, LH (1980) Lung crackles in bronchiectasis. Thorax 35,694-699[Abstract]
-
Shirai, F, Kudoh, S, Shibuya, A, et al (1981) Crackles in asbestos workers: auscultation and lung sound analysis. Br J Dis Chest 75,386-396[CrossRef][ISI][Medline]
-
Gilbert, VE (1989) Detection of pneumonia by auscultation of the lungs in the lateral decubitus positions. Am Rev Respir Dis 140,1012-1016[ISI][Medline]
-
Baughman, RP, Shipley, RT, Loudon, RG, et al (1991) Crackles in interstitial lung disease: comparison of sarcoidosis and fibrosing alveolitis. Chest 100,96-101[Abstract/Free Full Text]
-
Yasuda, N, Gotoh, K, Yagi, Y, et al (1997) Mechanism of posturally induced crackles as predictor of latent congestive heart failure. Respiration 64,336-341[ISI][Medline]
-
Murphy, RLJ, Del Bono, EA, Davidson, F (1989) Validation of an automatic crackle (rale) counter. Am Rev Respir Dis 140,1017-1020[ISI][Medline]
-
Ono, M, Arakawa, K, Mori, M, et al (1989) Separation of fine crackles from vesicular sounds by a nonlinear digital filter. IEEE Trans Biomed Eng 36,286-291[CrossRef][ISI][Medline]
-
Kaisla, T, Sovijarvi, A, Piirila, P, et al (1991) Validated method for automatic detection of lung sound crackles. Med Biol Eng Comput 29,517-521[CrossRef][ISI][Medline]
-
Arakawa, K, Harashima, H, Ono, M, et al (1991) Non-linear digital filters for extracting crackles from lung sounds. Front Med Biol Eng 3,245-257[Medline]
-
Sankur, B, Cagatay Guler, E, Kahya, YP (1996) Multiresolution biological transient extraction applied to respiratory crackles Comput Biol Med 26,25-39[CrossRef][ISI][Medline]
-
Celebi, G, Kalayci, T, Aysan, T, et al (1996) Application of multivariate linear discriminant analysis to lung sounds in some pulmonary diseases. Monaldi Arch Chest Dis 51,42-49[Medline]
-
Hadjileontiadis, LJ, Panas, SM (1997) Separation of discontinuous adventitious sounds from vesicular sounds using a wavelet-based filter. IEEE Trans Biomed Eng 44,1269-1281[CrossRef][ISI][Medline]
-
Vannuccini, L, Rossi, M, Pasquali, G (1998) A new method to detect crackles in respiratory sounds. Technol Health Care 6,75-79[Medline]
-
Sovijarvi, AR, Helisto, P, Malmberg, LP, et al (1998) A new versatile PC-based lung sound analyzer with automatic crackle analysis (HeLSA): repeatability of spectral parameters and sound amplitude in healthy subjects. Technol Health Care 6,11-22[Medline]
-
Pesu, L, Helisto, P, Ademovic, E, et al (1998) Classification of respiratory sounds based on wavelet packet decomposition and learning vector quantization. Technol Health Care 6,65-74[Medline]
-
Moore, BCJ (1992) An introduction to the psychology of hearing 3rd ed. ,106-119 Academic Press San Diego, CA.
-
Conijn, EA, Brocaar, MP, van Zanten, GA, et al (1992) Comparison between the frequency specificities of auditory brainstem response thresholds to clicks with and without high-pass masking noise. Audiology 31,284-292[ISI][Medline]
-
Owen, GA, Burkard, R (1991) Ipsilateral, contralateral, and binaural masking effects on the human brain-stem auditory-evoked responses to click stimuli. J Acoust Soc Am 89,1760-1767[CrossRef][ISI][Medline]
-
Holford SK. Discontinuous adventitious lung sounds: measurement, classification and modeling. PhD thesis, Massachusetts Institute of Technology, 1981; 69. Available at http://owens.mit.edu:8000/. Accessed February 22, 2001
-
Hoevers, J, Loudon, RG (1990) Measuring crackles. Chest 98,1240-1243[Abstract/Free Full Text]
-
Munakata, M, Ukita, H, Doi, I, et al (1991) Spectral and waveform characteristics of fine and coarse crackles. Thorax 46,651-657[Abstract]
-
Pasterkamp, H, Kraman, SS, DeFrain, PD, et al (1993) Measurement of respiratory acoustical signals: comparison of sensors. Chest 104,1518-1525[Abstract/Free Full Text]
-
Landis, JR, Koch, GG (1977) The measurement of observer agreement for categorical data. Biometrics 33,159-174[CrossRef][ISI][Medline]
-
Giard, MH, Fort, A, Mouchetant-Rostaing, Y, et al (2000) Neurophysiological mechanisms of auditory selective attention in humans. Front Biosci 5,D84-D94[ISI][Medline]
-
Pasterkamp, H, Powell, RE, Sanchez, I (1996) Lung sound spectra at standardized air flow in normal infants, children, and adults. Am J Respir Crit Care Med 154,424-430[Abstract]
-
Godfrey, S, Edwards, RH, Campbell, EJ, et al (1970) Clinical and physiological associations of some physical signs observed in patients with chronic airways obstruction. Thorax 25,285-287[ISI][Medline]
-
Brooks, D, Thomas, J (1995) Interrater reliability of auscultation of breath sounds among physical therapists. Phys Ther 75,1082-1088[Abstract/Free Full Text]
-
Wipf, JE, Lipsky, BA, Hirschmann, JV, et al (1999) Diagnosing pneumonia by physical examination: relevant or relic? Arch Intern Med 159,1082-1087[Abstract/Free Full Text]
-
Workum, P, DelBono, EA, Holford, SK, et al (1986) Observer agreement, chest auscultation, and crackles in asbestos-exposed workers. Chest 89,27-29[Abstract/Free Full Text]
-
Al Jarad, N, Strickland, B, Bothamley, G, et al (1993) Diagnosis of asbestosis by a time expanded wave form analysis, auscultation and high resolution computed tomography: a comparative study Thorax 48,347-353[Abstract]
-
Vandershoot, J, Helisto, P, Lipponen, P, et al (1998) Distribution of crackles on the flow-volume plane in different pulmonary diseases Technol Health Care 6,81-89[Medline]
-
Rossi, M, Vannuccini, L (1998) Placing crackles on the flow-volume plane: a study of the relationship between the time position, the flow and the volume Technol Health Care 6,91-97\[Medline]
TOP
Abstract
Introduction
