Download Respiratory Wheeze Sound Analysis Using Digital Signal

7th International Conference on Computational Intelligence, Communication Systems and Networks (CICSyN)
Respiratory Wheeze Sound Analysis Using Digital Signal Processing Techniques
Radwa Magdy Rady1*, Ibrahim Mohamed El Akkary2*+, Ahmed Nashaat Haroun3*+, Nader Abd Elmoneum
Fasseh4*, Mohamed Moustafa Azmy5*+
1
4
Bachelor of Engineering, Electronic and communication
department
2
professor, Department of Human Physiology
3
Emeritus Lecturer, Department of Biomedical
Engineering
present paper wheezes will be quantified and qualified. This
may help the physician to locate the place of obstruction and
its severity due to the features of quantifying and qualifying of
wheezes. Literature reports wide frequency range for wheezes.
The old ATS defined wheeze as having the pitch of < 400 Hz
and the duration of 250 msec. Subsequently, the ATS
Committee described wheeze occurring in adults, older
children and infants as having duration of > 250 msec. Also
the predominant peak frequency for wheeze was described to
be 225 Hz for infants and • 400 Hz for adults and older
children. The virtual auscultation CD (Compact Disk) by 3M
Littman mentions the duration of wheezes as • 200msec and
the dominant frequency of wheeze as • 300 Hz for adults and
225 Hz for infants. On the other hand the predominant energy
of normal lung sound lies in the frequency range of 100-200
Hz. There is neither musical character in the normal lung
sound nor distinct peaks above 200 Hz [20] [21].
Abstract— Auscultation and interpretation of lung sounds by
a stethoscope had been an essential method of diagnosing
pulmonary diseases. However this method has always been
unreliable due to poor audibility, inter-observer variations
(between different physicians). Thus computerized analysis of
lung sounds for diagnosis of pulmonary diseases is seen as a
convenient method. In the present paper different lung sounds
have been analyzed for wheeze detection and classification to
Monophonic or Polyphonic using MATLAB(Matrix Laboratory
software).The presented algorithm integrates and analyses the set
of parameters based on ATS (American Thoracic Society)
definition of wheeze and the previous researches. It is very
robust, computationally simple and yielded overall sensitivity of
90% for wheeze episode detection and accuracy of 91%. The
algorithm differentiates between monophonic wheezes and
polyphonic wheezes with sensitivity of 91% and accuracy of 70%.
In case of other lung sounds the proposed algorithm excluded
normal sounds from being identified as a wheeze with the
specificity of 90%.
Keywords— Wheeze properties; Respiratory Sounds; Digital
Signal Processing
I.
II.
SUBJECTS AND METHODS
The methods are examined on different lung sounds which
could be normal or adventitious. Anonymously lung sounds
are collected from two sources. Public repositories [22] [23]
[24] that are provided by online database, consist of. 63
different lung sounds. Private repository from a clinic of
pediatrician contains 143 lung sound files collected
electronically by using stethoscope (3M Littman Electronic
Stethoscope Model 3200) via Bluetooth.
The proposed approach detects and characterizes the
wheezes in lung sounds by using the MATLAB. The approach
consists of two stages. The first stage is respiration stage
consisting of the respiration detector, the respiratory rate
detector and the respiratory phase onset detector. The second
stage is the wheeze stage consisting of the wheeze detector,
quantifier and qualifier.
Signal transformation block is used before the respiration
stage for two goals. The first goal is obtaining signal
representation form that is useful for feature extraction in the
respiration stage, and the second goal is removing random
noise spikes in the input waveform as shown in ͞&ŝŐ͘ϭ͟.
INTRODUCTION
The importance of the paper is based on the difficulty of
examining patients, especially pediatrics, making it even more
difficult to properly diagnose the condition and execute the
treatment. The ability to differentiate between the numerous
types of respiratory sounds, especially wheezes, can be one of
the most important skills perfected by the physician.
Understanding airway anatomy and distinguishing between
upper and lower airway noises is sometimes a challenge [1].
Electronic auscultation; recording of lung sounds during
normal breathing and the analysis of the lung sounds after
transferring them to the computer is considered as reliable,
portable, simple, non-invasive and inexpensive technique [2].
From reference [3] to reference [19] are researches which
have been done relatively to the present technique that guides
us in the paper.
The brief review of those previous papers clarified that the
detection of wheezes and the study of some of their features
are not enough to diagnose the pulmonary pathology. In the
978-1-4673-7016-5/15 $31.00 © 2015 IEEE
DOI 10.1109/CICSyN.2015.38
professor, Department of pediatrics, Faculty of Medicine
5
Lecturer, Department of Biomedical Engineering
+
Medical Research Institute,
*Alexandria University
Alexandria, Egypt
146
162
The respiration detector detects regions in the acoustic
signal over which breathing is present. There the respiratory
rate detector computes the respiratory rate per minute and the
respiratory phase onset detector detects the onset-times of
each respiratory phase shown in “Fig.”2.
region of interest as a percentage. The duration of that region
is shown in “Fig.3”.
Fig.1.Time vs. Magnitude of the input signal, with an approximate curve
manually fitted to its envelope which is the output of the signal transformation
block.
Fig.3. Block diagram of the wheezes analysis system.
III.
RESULTS
The lung sound samples analyzed by the proposed
algorithm include inspiration, expiration, monophonic,
polyphonic, sibilant and sonorous wheezes. To establish the
robustness of the algorithm to rule out false positives, other
lung sounds are also analyzed by the same algorithm. They
include normal, abnormal and other adventitious lung sounds.
The results of the proposed method for wheeze file were saved
as JPEG (Joint Photographic Experts Group) images, which
are shown in “Fig.4” and “Fig.5”.
Fig.2.Block diagram of system design for detecting respiration, respiratory
rate, and respiratory phase onset.
The wheeze detector employs a time-dependent frequency
analysis on the signal. The estimated power spectral density is
done using auto regressive model by order 100. The method
looks for a prominent and isolated peak in the estimated power
spectral density of the signal by applying a series of threshold
tests to all detected peaks in the power spectrum that lie within
a specific frequency range. The wheeze quantifier detects the
total number of wheezes and their onset times. The wheeze
qualifier calculates the duration, mean frequency, and
frequency trend of each wheeze epoch. Each wheeze epoch
consists of sequential wheeze frames, where each wheeze
frame equals 300msec. It also classifies each wheeze epoch as:
monophonic (same pitch all over the frames or single frame in
the epoch), polyphonic (different pitches all over the wheeze
epoch), sibilant (high pitch mean frequency) or sonorous (low
pitch mean frequency) according to the fundamental
frequency of each wheeze frame that was detected as valid
peak in the wheeze detector block. It also calculates the
duration of total wheezing present in the signal over the entire
Fig.4. Analysis of the respiratory sound: Green: The signal vs. Time (in
seconds) to know its breathing intervals, the respiratory rate and the onsets of
the phases of respiration incase of wheeze sound.
163
147
Fig.7. Percentage of group A patients are above two years old, while group B
patients are below or equal two years old.
Fig.5. Wheeze analysis of wheeze sound: 1) Yellow: The signal, Red: The
analysis where 1(wheeze) and -1(no wheeze) vs. Time (in seconds)
2) Frequency trend vs. Time (in seconds)
Table I shows the results of the proposed approach on the
World Wide Web data. The results of the approach on the data
collected from the private clinic are classified according to the
patients’ sex, age and respiratory lung sounds, as shown in
“Fig.6”, “Fig.7” and “Fig.8”.
Fig.8. Lung sound classification according to the physician diagnosis.
TABLE I. RESULTS OF THE PROPOSED METHOD ON THE WORLD WIDE WEB
DATA
Types of Lung Sounds
I
II
Types of Wheeze
Sounds
Monophonic wheeze
No. of
Samples
Detected Correctly
26
21
8
7
Polyphonic wheeze
18
14
Other Lung Sounds
37
35
Normal Sounds
Other Adventitious
Sounds
13
13
24
22
The overall program sensitivity and accuracy for wheeze
episode detection are 90% and 91% respectively. The
algorithm differentiates between monophonic wheezes and
polyphonic wheezes with sensitivity of 91% and accuracy of
70%.
IV.
DICUSSION
In the presented research wheezes in the lung can be
detected as obstruction in the airways and its location may be
specified by understanding the different types of wheezes. The
sibilant wheeze is high pitch coming from narrow airway tube
or severe obstruction. The sonorous wheeze is low pitch
coming from wide airway tube or mild obstruction. The
monophonic wheeze is single-tone coming from one airway
tube and the polyphonic is multi-tone coming from more than
one closely located. The inspiratory wheeze indicates upper
airway obstruction and the expiratory wheeze indicates lower
airway obstruction due to the airflow mechanism of
respiration. The pathology of the patient can be diagnosed by
three types of wheeze: inspiratory or expiratory, sonorous or
sibilant and monophonic or polyphonic.
Fig.6. Percentage of male and female patients.
164
148
[7]
Continuous adventitious breath sounds are wheezes
rhonchi and stridor. Stridor is most commonly a result of
obstruction in the portions of the airway which are outside of
the chest cavity. It is an important distinction, since
obstruction of the airway within the chest cavity usually
presents as wheezing.
The results of the presented research show: a higher
accuracy in patients above two years of age, the sensitivity in
detecting monophonic wheezes higher than polyphonic
wheezes and 93% of the stridor sounds were identified as
monophonic wheeze.
So far, other researchers like Hashemi and his colleagues
reached 89% of accuracy in classifying wheezes to
monophonic and polyphonic [13]. They used the MLP neural
network as a classifier on the COPD and asthma subjects only.
Their problem is the computational complexity of the artificial
neural networks and the absence of normal and other different
adventious subjects. The present research solves this problem
by classifying the wheezes simply using given set of
parameters based on the ATS definition of wheeze. Emanet
and others found that advanced analytical techniques are
gaining popularity in many fields including healthcare and
medicine [18]. Their classification reached about 90% of
accuracy on predicting asthma. This shows that there is
possibility to increase the accuracy of the presented research.
This increase occurs by using two different methods. The first
is increasing the number of normal and adventious subjects.
The second is using spirometry and bronchoscope to validate
the presented program.
V.
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
CONCLUSION
Physicians can use this system to help them in detection of
the lung obstructive disease and it may specify the location of
the obstruction in the lung according to the analysis of lung
sounds. All they need to download the MATLAB complier
and run the executable file of the research program.
[19]
[20]
[21]
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
R. J. Leo, “Competency and the Capacity to Make Treatment Decisions:
A Primer for Primary Care Physicians,” Prim. Care. Companion J. Clin.
Psychiatry. 1st ed., vol. 1, pp. 131–141, oct 1999.
A. Marques, A. Bruton, A. Barney, “Clinically useful outcome measures
for physiotherapy airway clearance techniques: a review. Physical
Thearpy Reviews,” 11th ed., vol. 4, pp. 299–307, 2006.
doi:10.1179/108331906X163441
A. R. A. Sovijarvi, J. Vanderschoot, J. E. Earis, “Standardization of
computerized respiratory sound analysis,” Eur. Respir. Rev. 10th ed.,
vol. 77, pp. 585, 2000.
B. K. Serhat, A. Ertuzun, Y. P. Kahya, “Analysis and Classification of
Respiratory Sounds by Signal Coherence Method,” Proceedings of the
25thAnnual International Conference of the IEEE EMBS Cancun,
Mexico, vol.3, pp.2950-2953, 2003.
R. Jané, S. Cortés, J. A. Fiz, J. Morera, “Analysis of Wheezes in
Asthmatic Patients during Spontaneous Respiration,” Proceedings of the
26th Annual International Conference of the IEEE EMBS San Francisco,
CA, USA, vol.5, pp. 3836-3839, 2004.
Z. Moussavi, “Fundamentals of Respiratory Sounds and Analysis,” M &
C. BME . 1st ed., vol.1, pp. 1-68, 2006.
[22]
[23]
[24]
165
149
S. Reichert, R. Gass, C. Brandt, E. Andrès, “Analysis of Respiratory
Sounds: State of the Art,” Clin. Med. Circul. Respir. Pulm. Med., 1st
ed., vol.2, pp. 45-58, 2008.
A. Jain, J. Vepa. , “Lung Sound Analysis for Wheeze Episode
Detection,” 30th Annual International IEEE EMBS Conference
Vancouver, British Columbia, Canada, pp. 2582-2585, 2008.
S. Aydore, I. Sen, Y. P. Kahya, M. Mihcak,” Classification of
Respiratory Signals by Linear Analysis,” 31st Annual International
Conference of the IEEE EMBS Minneapolis, Minnesota, USA, pp.26172620, 2009.
A. Abbas, A. Fahim, “An Automated Computerized Auscultation and
Diagnostic System for Pulmonary Diseases,” J. Med. Syst., vol.34,
pp.1149-1155, 2010.
S. A. Taplidou, L. J. Hadjileontiadis, “Analysis of Wheezes using
Wavelet Higher Order Spectral Features,” IEEE Trans. Biomed. Eng.,
57th ed., vol.7, pp. 1596-1610, 2010.
B. Flietstra, N. Markuzon, A. Vyshedskiy, R. Murphy, “Automated
Analysis of Crackles in Patients with Interstitial Pulmonary Fibrosis,”
Pulmonary Medicine, 2011. doi.org/10.1155/2011/590506
A. Hashemi, H. Arabalibiek, K. Agin, “Classification of Wheeze Sounds
Using Wavelets and Neural Networks,” IPCBEE , pp.127-131, 2011.
L. E. Ellington, R. H. Gilman, J. M. Tielsch, M. Steinhoff, D. Figueroa,
S. Rodriguez, B. Caffo, B. Tracey, M. Elhilali, J. West, W. Checkley
,”Computerized lung sound analysis to improve the specificity of
pediatrics pneumonia diagnosis in resource-poor settings: protocol and
methods for an observational study,” BMJ Open, 2012.
E. Dimitra, P. Kailash, W. James, E. Mounya, “A multiresolution
analysis for detection of abnormal lung sounds,” 34th Annual
International Conference of the IEEE EMBS San Diego, California
USA, pp.3139-3142, 2012.
L. S. Morillo, S. A. Moreno, M. A. F. Granero, A. L. Jimcennz,
“Computerized analysis of respiratory sounds during COPD
exacerbations,” Comp. Bio. Med., 43rd ed., vol.7, pp.914-921, 2013.
N. Emanet, H. R. Öz, N. Bayram, D. Delen , “A comparative analysis of
machine learning methods for classification type decision problems in
healthcare,” Decision Analytics, 1st ed., vol.6, pp.1-20, 2014.
ȿ. G. Furman, ȿ. V. Yakovleva, S. V. Malinin, G. Furman, V.
Sokolovsky,“Computer-Assisted Assay of Respiratory Sounds of
Children Suffering from Bronchial Asthma,” ɋɌɆ , 6th ed., vol.1,
pp.83-87, 2014.
D. Oletic, B. Arsenali, V. Bilas , “Low-Power Wearable Respiratory
Sound Sensing,” Sensors, vol.14, pp. 6535-6566, 2014.
American Thoracic Society Ad Hoc Committee on Pulmonary
Nomenclature. 1977. Updated nomenclature for membership reaction.
ATS News 3:5–6.
G. Charbonneau, E. Ademovic, B. M. G. Cheetham, L. P. Malmberg, J
.Vanderschoot, A. R. A. Sovijärvi, “Basic techniques for respiratory
sound analysis,” Eur. Respir. Rev., 10th ed., vol.77, pp.625-626, 2000.
Pulmonary breath Sounds. East Tennessee State University, November
2002.
J. J, Ward. R.A.L.E. Lung Sounds Demo. Med. RRT in Respiratory
Care, Canada 2005; 50(10):1385-8.
Easy Auscultation. Med Edu, LLC. Westborough, MA 01581, USA
2010.