ArticlePDF Available

Cough Sound Analysis for Pneumonia and Asthma Classification in Pediatric Population

Authors:
  • Department of Paediatric, Sardjito Hospital/Faculty of Medicine, Gadjah Mada University, Yogyakarta, Indonesia

Abstract and Figures

Pneumonia and asthma are the common diseases in pediatric population. The diseases share some similarities of symptoms that make them difficult to separate without the proper diagnostic tools. The majority of pneumonia cases occur in the third world countries wherein even the basic diagnostic tools (e.g.: x-ray) are extremely rare. In these countries, the WHO recommends using rapid breathing and chest in-drawing as approach to diagnose pneumonia in children with cough. As the results, many asthma patients were misdiagnosed as pneumonia and prescribed for unnecessary antibiotic treatment. In this study, we propose a cough sound analysis based method to differentiate pneumonia from asthma. Cough is the major symptom of pneumonia and asthma. Past studies showed the acoustic of cough sounds may carry important information related with the diseases. However, there were no attempts to use cough sounds to separate pneumonia and asthma in pediatric population. Our method extracted sound features such as Mel-frequency cepstral coefficients, non-Gaussianity score and Shannon entropy. The features were then used to develop artificial neural network classifiers. Tested using leave one out validation technique in eighteen subjects, our method achieved sensitivity, specificity and Kappa of 89%, 100%, and 0.89 respectively. The results show the potential of our method to be developed as a tool to differentiate pneumonia from asthma in remote areas.
Content may be subject to copyright.
Cough Sound Analysis for Pneumonia and Asthma Classification in Pediatric
Population
Yusuf Amrulloh*, Udantha Abeyratne*, Vinayak Swarnkar*, Rina Triasih
*School of ITEE, The University of Queensland, Brisbane, Australia
Department of Electrical Engineering, Universitas Islam Indonesia, Yogyakarta, Indonesia
Department of Child Health, Sardjito Hospital, Universitas Gadjah Mada, Yogyakarta, Indonesia
e-mail: udantha@itee.uq.edu, yusuf.amrulloh@uii.ac.id
AbstractPneumonia and asthma are the common diseases in
pediatric population. The diseases share some similarities of
symptoms that make them difficult to separate without the
proper diagnostic tools. The majority of pneumonia cases occur
in the third world countries wherein even the basic diagnostic
tools (e.g.: x-ray) are extremely rare. In these countries, the
WHO recommends using rapid breathing and chest in-drawing
as approach to diagnose pneumonia in children with cough. As
the results, many asthma patients were misdiagnosed as
pneumonia and prescribed for unnecessary antibiotic
treatment. In this study, we propose a cough sound analysis
based method to differentiate pneumonia from asthma. Cough
is the major symptom of pneumonia and asthma. Past studies
showed the acoustic of cough sounds may carry important
information related with the diseases. However, there were no
attempts to use cough sounds to separate pneumonia and
asthma in pediatric population. Our method extracted sound
features such as Mel-frequency cepstral coefficients, non-
Gaussianity score and Shannon entropy. The features were
then used to develop artificial neural network classifiers. Tested
using leave one out validation technique in eighteen subjects,
our method achieved sensitivity, specificity and Kappa of 89%,
100%, and 0.89 respectively. The results show the potential of
our method to be developed as a tool to differentiate
pneumonia from asthma in remote areas.
Keywords-cough sound analysis, neural network, pediatrics,
pneumonia, asthma.
I. INTRODUCTION
Pneumonia is a serious threat for children, especially
those who live in the third world countries. In the population
of children younger than five years, there were around 120
million of pneumonia cases [1]. It was estimated that around
1.3 million children in that age died due to pneumonia every
year. Around 97% of pneumonia cases occurred in
developing countries and 74% of the cases occurred in south
Asia and sub-Saharan regions [2].
In similar, asthma is one of the most common chronic
respiratory diseases in pediatric population [3]. In United
States, around 14% of children admitted to hospital were
diagnosed with asthma [4]. The cost for pediatric asthma
treatment is estimated over $3 billion per year [5].
Pneumonia and asthma share similar symptoms such as
the difficulty of breathing and cough. In developed
countries, pneumonia be diagnosed more accurately using
imaging devices such as X-ray, CT-scan and supported by
blood tests as well as culture tests; whereas asthma by lung
function tests [5]. However, those diagnostic tools are not
readily available in the primary level of health care in third
world countries.
To accommodate diseases management in third world
countries, the World Health Organization (WHO) has
developed guidelines on Integrated Management of
Childhood Illnesses (IMCI) [6]. It contains the procedure for
diagnosing pneumonia and asthma in pediatric population.
According to that guideline, the clinical signs of rapid
breathing (respiratory rate greater than 50/min in children
younger than 12 months and greater than 40/min in children
older than 12 months to 60 months) in children with cough
should be treated as pneumonia and prescribed antibiotics.
The existence of lower chest in-drawing indicates the severe
pneumonia.
The evaluation on IMCI implementation in the field
showed that the guideline has relatively high sensitivity (69-
94%) but low specificity (16-67%) [7-9]. This means many
non-pneumonia children were misdiagnosed as pneumonia.
As the consequences, the children received unnecessary
antibiotics treatments. The misdiagnosis occurred because
the symptoms used to screen pneumonia also exist in
asthma. Recently, a study in Uganda [10] showed that 95%
of 253 children with asthma received antibiotics treatments
meant for pneumonia. The similar results were reported in
India [11] where 46% of 200 children with diagnosed as
pneumonia actually had asthma.
Researchers have attempted to improve the IMCI
guideline by augmenting extra symptoms such as fever and
nasal flaring [12-14]. However, the augmentation of
symptoms complicates the guidelines and requires skilled
health workers for the implementation. An alternative
method for screening pneumonia in remote areas is urgently
required.
Cough is one of the major symptoms of pneumonia as
well as asthma. Cough sound is believed carried information
2015 6th International Conference on Intelligent Systems, Modelling and Simulation
2166-0670/15 $31.00 © 2015 IEEE
DOI 10.1109/ISMS.2015.41
127
related to diseases. The study in [15] showed that the
wavelet coefficients extracted from the voluntary cough
sounds capable of differentiating groups of healthy,
asthmatic, and chronic obstructive pulmonary diseases
(COPD) subjects with accuracy 85-90%. However, the study
only included adult subjects and excluded pneumonia. Our
recent study [16] showed the plausibility of cough sound
analysis to differentiate pneumonia and non-pneumonia
diseases.
In this paper, we proposed an artificial neural network
(ANN) based classifier for pneumonia-asthma classification
using cough sounds. To the best of our knowledge, the work
on pneumonia-asthma classification using cough sounds is
the first effort in this field. As the novelty of this paper, it
contributes the following:
This paper addresses the basic problem of pneumonia
and asthma misdiagnosis, especially in the primary
level of health care in third world countries, where
experienced physicians and diagnosis tools are
extremely rare.
It demonstrates the use of cough sound analysis as a
novel tool to classify pneumonia from asthma.
The outcome of this study is significantly useful to
support the existing WHO guideline in pneumonia/asthma
management.
II. MATERIALS AND METHOD
A. Data acquisition
The data for this work were recorded at Sardjito
Hospital, Yogyakarta, Indonesia, from pediatric patients
admitted on respiratory complaints. The inclusion criteria
used in the recruitment was patients with at least two of the
following symptoms: cough, sputum, breathlessness, and
temperature higher than 37.5°C. We excluded patients
having advanced disease where recovery is not expected,
diseases with droplet precautions and patients undergoing
mechanical ventilation treatment. The recordings were
started after physicians had examined the subjects, begun the
initial treatment, and informed consent had been completed.
The duration of recording for each subject was from 1 6
hours depending on the condition of the patients. The
research protocol had received ethics clearances from
Sardjito Hospital and The University of Queensland,
Australia.
The data acquisition system consisted of a low-noise
microphone (Model NT3, RODE®, Sydney, Australia),
followed by a pre-amplifier and an A/D converter (Model
Mobile Pre-USB, M-Audio®, CA, USA). The output of the
Mobile Pre-USB was connected to the USB port of a laptop
computer. The nominal distance from the microphones to the
mouth of subjects was 50 cm. The actual distance could vary
from 40 cm to 100 cm due to the subject movement. We
kept the sampling rate at 44.1 k samples/s and 16-bit
resolution to obtain the best sound quality.
B. Construction of cough dataset
In this study, we involved M pediatric subjects (M = 18)
admitted to hospital with respiratory complaint. In the
dataset, the ratio of pneumonia and asthma subjects was
equal (9 of each disease). The clinical diagnosis of
pneumonia/asthma was established by the professional
pediatricians from Sardjito Hospital Yogyakarta Indonesia.
The cough dataset used in this study were constructed
by manually picking W first coughs (W = 50) from each
recording. The criteria of the cough selection were: i) cough
signal were not overlapped with other sounds and ii) cough
signal were not clipped. If the number of cough in a
recording less than W, then the maximum number of cough
that fulfilled the criteria were used. These coughs were used
to form cough dataset D. This dataset was then used to train
and test the artificial neural network (ANN) for
pneumonia/asthma classification. Details of this process are
described in the following section.
C. Classification of pneumonia and asthma
To classify pneumonia and asthma subjects we process
the cough episodes through three steps (S1-S3) as illustrated
in Fig 1. The steps are as follows:
Cough episodes s[n]
Reduce noise using high pass filter and power spectral
subtraction filter
(S1) Noise reduction
(S2) Feature extraction
Compute features in 20 ms sub-block
MFCC
Formant
Frequency
Shannon
Entropy
ZCR
NGS
Feature vectors of sub-blocks (F) from cough
episodes
R
 
|][
ˆ
| ..., |,][
ˆ
| ..., |,][
ˆ
| ][
ˆ1nsnsnsns K
k
Z
H
(S3)Pnemonia/asthma classification
Compute the performance of the ANN models
Train and test the ANN models following leave
one out validation method
Figure 1. Block diagram of the proposed method for
automatic pneumonia/asthma classification.
128
(S1) Noise reduction: Let s[n] denotes a discrete time signal
of a cough episode in dataset D. Process s[n] through high
pass filter (HPF) and power spectral subtractions (PSS)
filter. The HPF was designed as a fourth order Butterworth
filter with cut off frequency (fc = 10 Hz). The particular fc
was selected based on the low frequency noise profile in the
recordings due to the microphone stands vibration. The PSS
filter was used to reduce the Gaussian noise [17]. The
filtered cough episode was denoted by ŝ[n].
(S2) Feature extraction: We computed the feature vector of
each filtered cough episode ŝ[n]. The process of feature
vector follows the steps:
i. Apply a rectangular sliding window wr[n] of length N
(N = 882 samples, equal to 20 ms) to ŝ[n], generating
data sub-blocks. Let ŝ[n] = {| ŝ1[n]|, …, | ŝk[n]|, …, |
ŝK[n]|} represents the filtered sound recording where
ŝk[n] represents the kth (k = 1, 2, …, K) sub-block in
ŝ[n].
ii. For each sub-block ŝk[n] we computed the following
features:
Mel-frequency cepstral coefficients (MFCCs): The
MFCCs (Øk) of a sub-block ŝk[n] can be computed
using (1).
C
ckk C
cr
cL
12)12(
cos)(
(1)
where Lk is log energy output of c Mel Filter banks (c
= 1, 2, …, 40) of a sub-block ŝk[n] and r is the
number of cepstral coefficients (r = 0, 1, …, 12).
Formant frequency: In speech, formant frequency
shows the characteristics of vocal tract resonances.
We included the first five formant frequencies (R =
R1, R2, R3, R4, R5) in our feature set. We computed
the R1- R5 by peak picking the LPC spectrum. For
this work we used 14th order LPC spectrum and its
parameters were determined via Yule-Walker
autoregressive method along with the Levinson-
Durbin recursive procedure [18].
Zero crossing rate (ZCR): The ZCR (Zk), defined as
the total time a signal crosses the zero axis.
Non-Gaussianity score (NGS): The NGS provides an
easy method to quantify the deviation of a given
signal from a Gaussian model. The NGS (ψk) of a
sub-block ŝk[n] can be calculated using (2) [19],
where p and q are the normal probability plot of the
reference normal data and analyzed data,
respectively.
Nj
qq
pq
N
jj
N
jj
k
1 ,
)(
)(
1
1
2
1
2

Shannon entropy: The Shannon entropy (Hk) of a
sub-block ŝk[n] was obtained using definition in (3).
 
11 , )(
ˆ
ln)(
ˆ
1
1
22 N-nnsnsH N
nkkk

The feature vectors of ŝk[n] can be notated as Fk = [Ø
R
H]T. It comprised of 22 feature vectors.
iii. Repeat steps (i) and (ii) to all cough episodes in
dataset D and form feature vector matrix G (G = F1,
F2, …, FK).
In the next stage, the feature vector matrix G was processed
through the neural network classifier to differentiate
pneumonia to asthma subjects.
(S3) Classification of pneumonia and asthma using
neural network: To classify pneumonia and asthma
subjects, we designed an artificial neural network (ANN)
based classifier. The description of the classification
procedure and the ANN structure are as follows:
a. Classification procedure: Let Q = [Fk-2 Fk-1 Fk Fk+1
Fk+2], the element of feature vector matrix G, represents
the feature vectors of five successive sub-blocks ŝk-2[n],
ŝk-1[n], ŝk[n], ŝk+1[n], ŝk+2[n], respectively. In total, there
are 5 x 22 = 110 feature vectors in Q. We used Q as
input to the ANN and classified it into Pneumonia class
(“1”) or Asthma class (“0”). This process was repeated
for k+1, …, K to cover the whole signal ŝ[n] (ŝ[n] = {|
ŝ1[n]|, …, | ŝk[n]|, …, | ŝK[n]|}). The ANN was trained to
set the output to “1” for pneumonia class and “0” for
asthma. Let μo is the average of ANN output after
processing the sound features of all sub block of cough
Output: Pneumonia/Asthma Class
Output
layer
Hidden
layer
Input
layer
Neural network
F1
F2
...
...
Neurons
Neurons
Fk-2
Fk-1
Fk
Fk+1
Fk+2
Fk+3
... ...
FK-1
FK
Shifting
direction
Feature vectors
matrix
Figure 2. Illustration of ANN classification process. The
ANN is used to classify the sound features (F1, F2, FK)
computed from sub blocks of cough episodes (s1[n], s2[n],
sK[n]) into Pneumonia/Asthma class.
129
episodes from a patient. The classification of pneumonia
or asthma follows the rule:
if μo > γ then patient classified as pneumonia
if μo γ then patient classified as asthma
where γ is the optimized threshold.
b. ANN structure: The ANN structure comprised of an
input layer (Li), two hidden layers (Lh1 and Lh2), and an
output layer (Lo). The number of neurons in Li, Lh1, Lh2,
and Lo are 110, 20, 10, and 1 respectively. The numbers
of neurons in the input and output layers were designed
based on the number of feature vectors used as input for
the ANN and the required output. The neurons in hidden
layers were designed to achieve maximum performance
without over fitting. We used a linear activation
function for neurons in Lo layer and sigmoid activation
functions for neurons in Lh1 and Lh2 layers. To determine
initial weights and bias values, we used the Nguyen-
Widrow initialization method [20]. For updating
weights during the training process, we employed the
resilient back propagation (RPROP) algorithm [21].
We followed leave one out validation where all subjects
were used in training except one for testing. This
process was systematically repeated such that each
subject was used as the testing data once.
We show the neural network diagram and classification
procedure in Fig 2.
III. RESULTS
A. Dataset
In this study, we used recordings from M = 18 subjects
consisted of 7 male and 11 female. The age of the subjects
ranges from 1 86 months (average age = 25 months). Chest
x-ray was used to confirm 8 of the pneumonia subjects and 1
subject was clinically diagnosed pneumonia.
The total number of cough episodes in the data set D was
674. It consisted of 412 cough episodes from pneumonia
subjects and 262 cough episodes from asthma subject. The
average number of cough in pneumonia subjects is larger
than asthma subjects (41 coughs by 26 coughs, respectively).
The length of pneumonia coughs varied from 0.18 1.22 s
(mean = 0.33 s, median = 0.32 s and standard deviation =
0.12 s) while duration of asthma coughs varied from 0.2
1.44 s (mean = 0.49 s, median = 0.4 s and standard deviation
0.27 s). The results show that pneumonia coughs have
relatively shorter coughs than asthma.
According to physical examination findings, 8 of
pneumonia subjects and 5 of asthma subject had respiratory
rate above the threshold. Fever (body temperature > 37.5°C)
was presence in 6 pneumonia subjects and 4 asthma
subjects. Abnormal lung sounds found in these subjects were
crackles (pneumonia = 8 and asthma = 1) and wheeze
(pneumonia = 1 and asthma = 7). Respiratory distress
symptoms (sub costal retractions) were presence in all
pneumonia subjects and in 3 asthma subject.
B. Pneumonia/asthma classification
In Fig 3, we show the receiver operating characteristic
curve (ROC) of the ANN models used for
pneumonia/asthma classification. By optimizing the
sensitivity and specificity in the training set, we defined an
optimum threshold (γ = 0.6).
The results of pneumonia and asthma classification using
the optimum threshold γ are shown in Table 1. It can be seen,
the testing results generated from leave one out validation
show that the developed algorithm is capable of classifying
pneumonia and asthma with high sensitivity 88.9 % and
specificity 100%. The Kappa agreement with the diagnosis
from the professional pediatricians is also very high (0.89).
IV. DISCUSSION AND CONCLUSION
The physical examination findings show that more than
50% of asthma subjects had respiratory rate higher than
threshold and 30% of them had sub-costal retraction. It
means that if IMCI guidelines applied for diagnosing these
patients, they will be misclassified as pneumonia. Study in
[12] suggested adding fever to improve the specificity of
pneumonia diagnosis. However, 44.4% of asthma subjects
had fever. The physical examinations also show that crackles
sounds is not specific to pneumonia.
TABLE 1. The results of pneumonia/asthma classification
following leave one out validation technique. Tr, Te, γ, Sens,
Spec, Acc, PPV, NPV and κ, respectively denote training,
testing, optimized threshold, sensitivity, specificity,
accuracy, positive predictive value, negative predictive value
and Cohen’s Kappa statistic.
γ
Sens
Spec
Acc
PPV
NPV
κ
Tr
0.6
92.1
100.0
96.0
100.0
92.6
0.92
Te
0.6
88.9
100.0
94.4
100.0
90.0
0.89
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
1-Specificity
Sensitivity
Training
Testing
Figure 3. ROC curve of ANN used for pneumonia/asthma
classification.
130
In this paper we proposed an ANN based classifier to
differentiate pneumonia subjects with asthma subjects using
their cough sounds. Tested in 18 subjects following leave one
out validation technique, our method achieved high
sensitivity, specificity and Kappa. This result supports our
previous study [16] that cough sound carry important
information useful to screen respiratory diseases. Our study
shows that cough sound analysis has potential to be
developed as screening tool for differentiating pneumonia
from asthma in remote areas. However, these positive results
should be followed by study in the larger dataset.
REFERENCES
[1] C. L. F. Walker, I. Rudan, L. Liu, H. Nair, E. Theodoratou, Z. A.
Bhutta, et al., "Global burden of childhood pneumonia and diarrhoea,"
The Lancet, vol. 381, pp. 1405-1416, 2013.
[2] I. Rudan, C. Boschi-Pinto, Z. Biloglav, K. Mulholland, and H.
Campbell, "Epidemiology and etiology of childhood pneumonia,"
Bulletin of the World Health Organization, vol. 86, pp. 408-416, 2008.
[3] World Health Organization, "Chronic respiratory diseases," ed.
Geneva: World Health Organization, 2012.
[4] B. Bloom, R. A. Cohen, and G. Freeman, "Summary health statistics
for U.S. children: National Health Interview Survey, 2010," National
Center for health Statistics. Vital Health Statistics 10(250), 2011.
[5] L. J. Stewart, "Pediatric Asthma," Primary care, vol. 35, pp. 25-40,
2008.
[6] World Health Organization, "Handbook: IMCI integrated management
of childhood illness," ed. Geneva: World Health Organization and
UNICEF, 2005.
[7] M. Palafox, H. Guiscafré, H. Reyes, O. Munoz, and H. Martínez,
"Diagnostic value of tachypnoea in pneumonia defined
radiologically.," Arch Dis Child., vol. 82(1), pp. 41-5, 2000.
[8] M. Harari, V. Spooner, S. Meisner, M. Carney, F. Shann, and J. de
Campo, "Clinical signs of pneumonia in children," The Lancet, vol.
338, pp. 928-930, 1991.
[9] A. Falade, H. Tschappeler, B. M. Greenwood, and K. Mulholland,
"Use of simple clinical signs to predict pneumonia in young Gambian
children: the influence of malnutrition," Bull World Health Organ,
vol. 73:299, 1995.
[10] R. Nantanda, J. K. Tumwine, G. Ndeezi, and M. S. Ostergaard,
"Asthma and Pneumonia among Children Less Than Five Years with
Acute Respiratory Symptoms in Mulago Hospital, Uganda: Evidence
of Under-Diagnosis of Asthma," PLoS ONE, vol. 8, p. e81562, 2013.
[11] H. P. Sachdec, S. C. Mahajan, and A. Garg, "Improving antibiotic and
bronchodilator prescription in children presenting with difficult
breathing: experience from urban hospital in India," Indian Pediatr.,
vol. 38(8), pp. 827-38, 2001.
[12] M. A. Cardoso, C. M. Nascimento-Carvalho, F. Ferrero, F. M. Alves,
and S. N. Cousens, "Adding fever to WHO criteria for diagnosing
pneumonia enhances the ability to identify pneumonia cases among
wheezing children," Archives of Disease in Childhood, vol. 96, pp. 58-
61, 2011.
[13] A. V. Castro, C. M. Nascimento-Carvalho, F. Ney-Oliveria, C. A.
Araujo-Neto, S. C. Andrade, L. L. Loureiro, et al., "Additional
markers to refine the World Health Organization algoithm for
diagnosis of pneumonia," Indian Pediatr., vol. 42(8), pp. 773-81,
2005.
[14] S. M. Graham, M. English, T. Hazir, P. Enarson, and T. Duke,
"Challenges to improving case management of childhood pneumonia
at health facilities in resource-limited settings," Bull World Health
Organ., vol. 86(5), pp. 349-55, 2008.
[15] J. Knocikova, J. Korpas, M. Vrabec, and Javorka, "Wavelet analysis
of voluntary cough sound in patients with respiratory diseases,"
Journal of physiology and pharmacology : an official journal of the
Polish Physiological Society, vol. 59, pp. 331-40., 2009.
[16] U. R. Abeyratne, V. Swarnkar, R. Triasih, and A. Setyati, "Cough
sound analysis can rapidly diagnose childhood pneumonia," Annals of
Biomedical Engineering, vol. 41(11), pp. 2448-62, 2013.
[17] R. Martin, "Noise power spectral density estimation based on optimal
smoothing and minimum statistics," Speech and Audio Processing,
IEEE Transactions on, vol. 9, pp. 504-512, 2001.
[18] J. E. Markel and A. H. Gray, Linear prediction of speech. New York:
Springer-Verlag Inc, 1982.
[19] H. Ghaemmaghami, U. R. Abeyratne, and C. Hukins, "Normal
probability testing of snore signals for diagnosis of obstructive sleep
apnea," in Engineering in Medicine and Biology Society, 2009. EMBC
2009. Annual International Conference of the IEEE, 2009, pp. 5551-
5554.
[20] D. Nguyen and B. Widrow, "Improving the learning speed of 2-layer
neural networks by choosing initial values of the adaptive weights," in
IJCNN International Joint Conference on Neural Networks 1990, pp.
21-26 vol.3.
[21] M. Riedmiller and H. Braun, "A direct adaptive method for faster
backpropagation learning: the RPROP algorithm," in IEEE
International Conference on Neural Networks 1993, pp. 586-591
vol.1.
131
... The study showed that the overall sensitivity and specificity of machine learning classifier were 82.81% and 84.76%, respectively. In Ref. [31], Amrulloh et al. introduced a classification model for pneumonia and asthma. Their approach quantified the sound features by MFCC, Shannon entropy, and non-Gaussian, and these characteristics were found to form the basis for artificial neural network classifiers. ...
Article
Deep Learning shows promising performance in diverse fields and has become an emerging technology in Artificial Intelligence. Recent visual recognition is based on the ranking of photographs and the finding of artefacts in those images. The aim of this research is to classify the different cough sounds of COVID-19 artefacts in the signals of altered real-life environments. The introduced model takes into consideration two major steps. The first step is the transformation phase from sound to image that is optimized by the scalogram technique. The second step involves feature extraction and classification based on six deep transfer models (GoogleNet, ResNet18, ResNet50, ResNet101, MobileNetv2, and NasNetmobile). The dataset used contains 1457 (755 of COVID-19 and 702 of healthy) wave cough sounds. Although our recognition model performs the best, its accuracy only reaches 94.9% based on SGDM optimizer. The accuracy is promising enough for a wide set of labeled cough data to test the potential for generalization. The outcomes show that ResNet18 is the most stable model to classify the cough sounds from a limited dataset with a sensitivity of 94.44% and a specificity of 95.37%. Finally, a comparison of the research with a similar analysis is made. It is observed that the proposed model is more reliable and accurate than any current models. Cough research precision is promising enough to test the ability for extrapolation and generalization.
... Methods for automated cough-based diagnosis of respiratory diseases borrow many ideas and share technical similarities with detection. For instance, pneumonia can be discerned from other respiratory infections with a logistic regression on the acoustic features of a cough supplemented with wavelet transforms, [24], [25], or pertussis can be identified by a dedicated whoop detector and a cough classifier built with logistic regressions with greedily selected features, [19]. Chronic Obstructive Pulmonary Disease can be identified with a random forest, [26], or gradient boosted classifier ensembles on the MFCC features, [27]. ...
Article
The COVID-19 pandemic created a significant interest and demand for infection detection and monitoring solutions. In this paper we propose a machine learning method to quickly triage COVID-19 using recordings made on consumer devices. The approach combines signal processing methods with fine-tuned deep learning networks and provides for signal denoising, cough detection and classification. We have also developed and deployed a mobile application that uses symptoms checker together with voice, breath and cough signals to detect COVID-19 infection. The application showed robust performance on both open sourced datasets and on the noisy data collected during beta testing by the end users.
... The consideration of other features may result in more clinically applicable and accurate results. Many studies have shown that cough symptoms from multiple respiratory conditions have distinguishable features [14][15][16][17], COVID-19 has been noted to share similar latent cough traits as other respiratory conditions [18][19][20]. For instance, a patient with non-COVID-19-associated pneumonia may be falsely identified by a machine learning model as COVID positive due to the large prevalence of pneumonia that develops in severe COVID-positive patients [21]. ...
... They stated inexpensive devices like mobile apps, text messaging/SMS and wearable technologies had proved their feasibility in diagnosing different respiratory diseases. Also, Amrulloh et al. [6] reviewed the AI techniques that were used to detect asthma disease. The survey discussed the most used AI methods to detect asthma and the most used techniques are ANN (Artificial Neural Network), DT (Decision Tree) and RF (Random Forest). ...
Article
Full-text available
Coughing is a common symptom of several respiratory diseases. The sound and type of cough are useful features to consider when diagnosing a disease. Respiratory infections pose a significant risk to human lives worldwide as well as a significant economic downturn, particularly in countries with limited therapeutic resources. In this study we reviewed the latest proposed technologies that were used to control the impact of respiratory diseases. Artificial Intelligence (AI) is a promising technology that aids in data analysis and prediction of results, thereby ensuring people’s well-being. We conveyed that the cough symptom can be reliably used by AI algorithms to detect and diagnose different types of known diseases including pneumonia, pulmonary edema, asthma, tuberculosis (TB), COVID19, pertussis, and other respiratory diseases. We also identified different techniques that produced the best results for diagnosing respiratory disease using cough samples. This study presents the most recent challenges, solutions, and opportunities in respiratory disease detection and diagnosis, allowing practitioners and researchers to develop better techniques.
Article
Cough acoustics contain multitudes of vital information about pathomorphological alterations in the respiratory system. Reliable and accurate detection of cough events by investigating the underlying cough latent features and disease diagnosis can play an indispensable role in revitalizing the healthcare practices. The recent application of Artificial Intelligence (AI) and advances of ubiquitous computing for respiratory disease prediction has created an auspicious trend and myriad of future possibilities in the medical domain. In particular, there is an expeditiously emerging trend of Machine learning (ML) and Deep Learning (DL)-based diagnostic algorithms exploiting cough signatures. The enormous body of literature on cough-based AI algorithms demonstrate that these models can play a significant role for detecting the onset of a specific respiratory disease. However, it is pertinent to collect the information from all relevant studies in an exhaustive manner for the medical experts and AI scientists to analyze the decisive role of AI/ML. This survey offers a comprehensive overview of the cough data-driven ML/DL detection and preliminary diagnosis frameworks, along with a detailed list of significant features. We investigate the mechanism that causes cough and the latent cough features of the respiratory modalities. We also analyze the customized cough monitoring application, and their AI- powered recognition algorithms. Challenges and prospective future research directions to develop practical, robust, and ubiquitous solutions are also discussed in detail.
Article
Full-text available
Easy detection of COVID-19 is a challenge. Quick biological tests do not give enough accuracy. Success in the fight against new outbreaks depends not only on the efficiency of the tests used, but also on the cost, time elapsed and the number of tests that can be done massively. Our proposal provides a solution to this challenge. The main objective is to design a freely available, quick and efficient methodology for the automatic detection of COVID-19 in raw audio files. Our proposal is based on automated extraction of time-frequency cough features and selection of the more significant ones to be used to diagnose COVID-19 using a supervised machine-learning algorithm. Random Forest has performed better than the other models analysed in this study. An accuracy close to 90% was obtained. This study demonstrates the feasibility of the automatic diagnose of COVID-19 from coughs, and its applicability to detecting new outbreaks.
Article
Full-text available
Effective case management is an important strategy to reduce pneumonia-related morbidity and mortality in children. Guidelines based on sound evidence are available but are used variably. This review outlines current guidelines for childhood pneumonia management in the setting where most child pneumonia deaths occur and identifies challenges for improved management in a variety of settings and different "at-risk" groups. These include appropriate choice of antibiotic, clinical overlap with other conditions, prompt and appropriate referral for inpatient care, and management of treatment failure. Management of neonates, and of HIV-infected or severely malnourished children is more complicated. The influence of co-morbidities on pneumonia outcome means that pneumonia case management must be integrated within strategies to improve overall paediatric care. The greatest potential for reducing pneumonia-related deaths in health facilities is wider implementation of the current guidelines built around a few core activities: training, antibiotics and oxygen. This requires investment in human resources and in equipment for the optimal management of hypoxaemia. It is important to provide data from a variety of epidemiological settings for formal cost-effectiveness analyses. Improvements in the quality of case management of pneumonia can be a vehicle for overall improvements in child health-care practices.
Article
Full-text available
Pneumonia is considered the major cause of mortality among children with acute respiratory disease in low-income countries but may be over-diagnosed at the cost of under-diagnosing asthma. We report the magnitude of asthma and pneumonia among "under-fives" with cough and difficulty breathing, based on stringent clinical criteria. We also describe the treatment for children with acute respiratory symptoms in Mulago Hospital. We enrolled 614 children aged 2-59 months with cough and difficulty breathing. Interviews, physical examination, blood and radiological investigations were done. We defined asthma according to Global Initiative for Asthma guidelines. Pneumonia was defined according to World Health Organization guidelines, which were modified by including fever and white cell count, C-reactive protein, blood culture and chest x-ray. Children with asthma or bronchiolitis were collectively referred to as "asthma syndrome" due to challenges of differentiating the two conditions in young children. Three pediatricians reviewed each participant's case report post hoc and made a diagnosis according to the study criteria. Of the 614 children, 41.2% (95% CI: 37.3-45.2) had asthma syndrome, 27.2% (95% CI: 23.7-30.9) had bacterial pneumonia, 26.5% (95% CI: 23.1-30.2) had viral pneumonia, while 5.1% (95% CI: 3.5-7.1) had other diagnoses including tuberculosis. Only 9.5% of the children with asthma syndrome had been previously diagnosed as asthma. Of the 253 children with asthma syndrome, 95.3% (95% CI: 91.9-97.5) had a prescription for antibiotics, 87.7% (95% CI: 83.1-91.5) for bronchodilators and 43.1% (95% CI: 36.9-49.4) for steroids. Although reports indicate that acute respiratory symptoms in children are predominantly due to pneumonia, asthma syndrome contributes a significant proportion. Antibiotics are used irrationally due to misdiagnosis of asthma as pneumonia. There is need for better diagnostic tools for childhood asthma and pneumonia in Uganda.
Article
Full-text available
Pneumonia annually kills over 1,800,000 children throughout the world. The vast majority of these deaths occur in resource poor regions such as the sub-Saharan Africa and remote Asia. Prompt diagnosis and proper treatment are essential to prevent these unnecessary deaths. The reliable diagnosis of childhood pneumonia in remote regions is fraught with difficulties arising from the lack of field-deployable imaging and laboratory facilities as well as the scarcity of trained community healthcare workers. In this paper, we present a pioneering class of technology addressing both of these problems. Our approach is centred on the automated analysis of cough and respiratory sounds, collected via microphones that do not require physical contact with subjects. Cough is a cardinal symptom of pneumonia but the current clinical routines used in remote settings do not make use of coughs beyond noting its existence as a screening-in criterion. We hypothesized that cough carries vital information to diagnose pneumonia, and developed mathematical features and a pattern classifier system suited for the task. We collected cough sounds from 91 patients suspected of acute respiratory illness such as pneumonia, bronchiolitis and asthma. Non-contact microphones kept by the patient's bedside were used for data acquisition. We extracted features such as non-Gaussianity and Mel Cepstra from cough sounds and used them to train a Logistic Regression classifier. We used the clinical diagnosis provided by the paediatric respiratory clinician as the gold standard to train and validate our classifier. The methods proposed in this paper could separate pneumonia from other diseases at a sensitivity and specificity of 94 and 75% respectively, based on parameters extracted from cough sounds alone. The inclusion of other simple measurements such as the presence of fever further increased the performance. These results show that cough sounds indeed carry critical information on the lower respiratory tract, and can be used to diagnose pneumonia. The performance of our method is far superior to those of existing WHO clinical algorithms for resource-poor regions. To the best of our knowledge, this is the first attempt in the world to diagnose pneumonia in humans using cough sound analysis. Our method has the potential to revolutionize the management of childhood pneumonia in remote regions of the world.
Article
Objectives This report presents both age- adjusted and unadjusted statistics from the 2006 National Health Interview Survey (NHIS)on selected health measures for children under 18 years of age, classified by sex, age, race, Hispanic origin,family structure,parent education, family income, poverty status, health insurance coverage, place of residence, region, and current health status.The topics covered are asthma, allergies,learning disability, Attention Deficit Hyperactivity Disorder (ADHD), prescription medication use, respondent-assessed health status, school-loss days, usual place of health care, time since last contact with a health care professional, selected measures of health care access and utilization, and dental care. Source of Data The NHIS is a multistage probability sample survey conducted annually by interviewers of the U.S. Census Bureau for the Centers for Disease Control and Prevention's National Center for Health Statistics and is representative of the civilian noninstitutionalized population of the United States. Data are collected for all family members during face-to- face interviews with adults present at the time of interview. Additional information about children is collected for one randomly selected child per family in face-to-face interviews with an adult proxy respondent familiar with the child's health. Selected Highlights In 2006, most U.S. children under 18 years of age had excellent or very good health (82%). However, 10% of children had no health insurance coverage, and 5% of children had no usual place of health care. Fourteen percent of children had ever been diagnosed with asthma. An estimated 8% of children 3-17 years of age had a learning disability, and an estimated 7% of children had ADHD.
Article
La pneumonie est la principale cause simple de mortalite chez les enfants de moins de 5 ans. L'incidence dans cette tranche d'âge est estimee a 0,29 episode/enfant/an dans les pays en developpement et a 0,05 episode/enfant/an dans les pays developpes. Il en resulte environ 156 millions de nouveaux episodes de pneumonie chaque annee dans le monde, dont 151 millions dans les pays en developpement. La plupart des cas se produisent en Inde (43 millions), en Chine (21 millions), au Pakistan (10 millions) et egalement en grands nombres au Bengladesh, en Indonesie et au Nigeria (6 million pour chacun de ces pays). Parmi l'ensemble des cas communautaires, 7 a 13 % sont assez graves pour menacer le pronostic vital et necessiter une hospitalisation. De nombreux elements ont fait apparaitre comme facteurs de risque principaux pour l'incidence de la pneumonie l'absence d'allaitement au sein exclusif, la denutrition, la pollution de l'air interieur, le petit poids a la naissance, le surpeuplement et le manque de couverture par la vaccination antirougeoleuse. La pneumonie est responsable d'environ 19 % des deces d'enfants de moins de 5 ans, dont plus de 70 % se produisent en Afrique sub-saharienne et en Asie du Sud-est. Bien que reposant sur les donnees disponibles limitees, les etudes recentes ont identifie Streptococcus pneumonia, Haemophilus influenzae et le virus respiratoire syncytial comme les principaux agents pathogenes associes a la pneumonie de l'enfant.
Article
Diarrhoea and pneumonia are the leading infectious causes of childhood morbidity and mortality. We comprehensively reviewed the epidemiology of childhood diarrhoea and pneumonia in 2010-11 to inform the planning of integrated control programmes for both illnesses. We estimated that, in 2010, there were 1·731 billion episodes of diarrhoea (36 million of which progressed to severe episodes) and 120 million episodes of pneumonia (14 million of which progressed to severe episodes) in children younger than 5 years. We estimated that, in 2011, 700 000 episodes of diarrhoea and 1·3 million of pneumonia led to death. A high proportion of deaths occurs in the first 2 years of life in both diseases-72% for diarrhoea and 81% for pneumonia. The epidemiology of childhood diarrhoea and that of pneumonia overlap, which might be partly because of shared risk factors, such as undernutrition, suboptimum breastfeeding, and zinc deficiency. Rotavirus is the most common cause of vaccine-preventable severe diarrhoea (associated with 28% of cases), and Streptococcus pneumoniae (18·3%) of vaccine-preventable severe pneumonia. Morbidity and mortality from childhood pneumonia and diarrhoea are falling, but action is needed globally and at country level to accelerate the reduction.
Conference Paper
The authors describe how a two-layer neural network can approximate any nonlinear function by forming a union of piecewise linear segments. A method is given for picking initial weights for the network to decrease training time. The authors have used the method to initialize adaptive weights over a large number of different training problems and have achieved major improvements in learning speed in every case. The improvement is best when a large number of hidden units is used with a complicated desired response. The authors have used the method to train the truck-backer-upper and were able to decrease the training time from about two days to four hours