You are on page 1of 5
 
Volume 2, Issue 9, September 
 – 
 2017 International Journal of Innovative Science and Research Technology ISSN No: - 2456
 – 
 2165 IJISRT17SP137 www.ijisrt.com 283
An Integrated Computerized Cough Analysis by Using Wavelet for Pneumonia Diagnosis
Roshan S. Hande Pallavi S. Deshpande Department of Electronics & Telecommunication Depaartment of Electronics & Telecommunication Engineering, Smt. Kashibai Navale College of Engineering.Smt.Kashibai Navle College of Engineering, Vadgaon Bk. Pune. Engineering ,Vadgaon Bk.Pune. roshan9028037541@gmail.com  pallavisd@rediffmail.com
Abstract
-
Respiratory diseases such as pneumonia, bronchitis leading causes of child death in the word .out of this pneumonia are causing the million children death each year around the word. One of the challenged faced in consistent diagnosis of childhood pneumonia in secluded area is difficulties arising from field deployable, laboratory facilities and trained healthcare worker. Such issue we address in this paper and to categorize the pneumonia using the geometrical analysis of cough sound. We used the wavelet-based mathematical tool which is a useful work for crackle detection in lung sound analysis. Such feature can be added among new mathematical feature and to develop the automated classifier to distinguish the pneumonia with other respiratory diseases. In our project uses feed forward neural network classifier to increase the classification performance with having sensitivity 90%, specificity 98.7% and accuracy 97%.Cough and crackle sound are sign of pneumonia. Cough sounds permit us for pneumonia diagnosis with adequate sensitivity and specificity. Keywords:-
Slant Wavelet Transform, Neural Network, Pneumonia Cough Sample Sound.
I.
 
INTRODUCTION
Cough is a justification system to the body which clears the respiratory tract from outside materials which are inhale accidentally and create internally by infections. It is a common symptom appearing in respiratory diseases such as pneumonia, the foremost of death is occouring in children which is less than five years of age. It has been estimate that pneumonia shall causes over 1.5 million deaths in each year, with more Than 96% of cases occurring in the well developing countries. Main reason behind them is the facility which is available having low cost instrument, field-deployable and diagnostic technology is most challenges key in struggle pneumonia Mortality. Currently does not have special method or standard is an available for pneumonia diagnosis even in hospitals. [1], The process which is available is not simple, but rather a grouping of clinical, radiological, and laboratory diagnostics that is often difficult to get to much of the population affected  by the disease. Address such issue then developing an automated cough sound analysis method to diagnose  pneumonia. This Will possible to develop the system which has inexpensive, noncontact, way of testing pneumonia cases without the help for widespread training in the field. aim to  build a higher [1].specificity and maintain sensitivity at
>
90%.That study is a combination of several geometrical features, few of which are widely used in speech signal  processing, such as [4],formant frequencies (FF) and Mel Frequency Cepstral Coefficients (MFCC). Work shown in this  paper we intend the different class of features inspired by the adventitious lung sounds known as crackles, which is normally found in pneumonia and regularly observed more in the chest musculature using stethoscopes. We recorded cough sound signal with sound proof room in free-air outside the mouth and analyzed the (wavelet decomposition), targeting crackle-like components. We then combined the two feature sets and developed pattern recognition technology to diagnose childhood pneumonia. [5].Wavelets transform can provide a best way of resolve the nonstationary signals such as the crackle sound in both time and frequency domains. Wavelet having the capability to attention on restricted signal structures with a zooming  procedure is efficient in detect singularities between signals, and a powerful multiresolution analysis tool to detain changes in frequency characteristics at any instant in time. The diagnosis of childhood pneumonia using cough sound analysis is a like new research area. Our aim to explain the wavelets can be very effective in decomposing cough sounds and developing features definite to pneumonia.
 
Volume 2, Issue 9, September 
 – 
 2017 International Journal of Innovative Science and Research Technology ISSN No: - 2456
 – 
 2165 IJISRT17SP137 www.ijisrt.com 284
II.
 
OBJECTIVES
The Objective of this project listed below
 
Extract the feature of cough sound using the wavelet for diagnosis of pneumonia
 
To archive more accuracy of system
 
To make system more flexible and robust
III.
 
LITERATURE SURVEY
 [1]The paper by U. R. Abeyratne
 
in this research paper explains the cough sound analysis can be used to diagnose the child hood pneumonia. In this method the computerized study of cough signal and respiratory sound can be collected using microphone that does not require any direct contact with subject. Then segmentation had done using the manually from this find out mathematical feature, Such as non gaussianity and mel cepstra from cough sound. In this method differtiation of pneumonia and non pneumonia sound can done using logistic regression classifier
 [6]The paper by F. Ayari
 works going on in this paper show that lung sounds analysis can done using wavelet transform The objective of this paper for lung sounds analysis can be done using adaptive filtering and wavelets show with one desertion moment can successfully detect .the pathological changes of the lung which produce sounds with measurable regularities. Local regularity can allows us to detect some important components of adventitious sounds which are difficult to detect by the physician ears due to their short duration. to analyze lung sound it can uses the mathematical tool lipschitz continuity function which can detect the maxima position and minima position regular lung sound waveform pattern. Numerical results show that normal lung sound is not regular than as compare to the crackle lung sound
 [13]The paper by M. Du
 work going on this paper explain that Crackle sound classification and detection will based on matched wavelet analysis This is new method for crackle
detection which is depends upon the ‘matched’ wavelet
transform. Based on the Crackles sound can be detected using the envelope of the signal at optimal scale, and it can be classified based on energy distribution with scale.
 [4]The paper by vinayak swarnkar
 
“In this paper Automatic
segmentation of pneumonia cough and non-contact sound
recordings done in pediatric wards’’ In this paper developed a
method which can differentiate non pneumonia and  pneumonia cough segments automatically during the  pediatric sound recordings. Method is based on extracting statistical features such as non-Gaussianity, Shannon entropy, and mel frequency cepstral coefficients to describe cough characteristics. These features then used to train a time delay artificial neural network classifier to detect coughs segment in the sound recordings. From this proposed method achieve the sensitivity, specificity of 93%, 98%, respectively.
 [5]The paper by Yosuf Amrulloh, Rina Triasih
 in this research paper show that Pneumonia and asthma can be differtiate in pediatric Population based on cough sound analysis.This paper explains that Pneumonia and asthma are the common diseases in pediatric population. The diseases showing few similarities of symptoms that Cough is the major symptom of pneumonia and asthma. The audio of cough sounds may carry vital information which correlated with the diseases. This technique obtains the sound features such as Shannon entropy, mel frequency cepestral coefficient,  bispectrum score and kurtosis. This features then used to develop artificial neural network classifiers. [4].Using this classifier achieved specificity, Kappa and sensitivity of 100%, 0.89 and 89% respectively. 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. Study in 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. .
IV.
 
BACKGROUND
 A.
 
Continuous Wavelet Transform
The continuous wavelet transform uses signal and an analyzing function .it is different approach for simultaneous find out time and frequency signal. Wavelet has the advantage That it allows superior perceptible localization of frequency component to analyzed signal than commonly used short time Fourier transforms (STFT). Wavelet analysis allows to use long time windows function when we need the more specific low frequency signal. It can produce the exact representation for nonstationary signals with discontinuities like cough and crackle sounds. [4].The continues wavelet transform is given  by
CWTxi (a, b) =
xi, ψ a,b
=
 ʃ 
 
xiψa,bdt
 
Where
a
is the dilation parameter and
b
is the translation. The Dilation parameter is alike to the scale, which determines the timescale resolution of the resulting CWT operation. By analyzing
 xi
over a different range of scales, CWT offers multiresolution frequency filtering capability to target specific frequency bands. This change to dissimilar crackle types (coarse and fine) based on two cycle duration (2CD) of the detected crackles. Fig. 2 shows a time-domain example of an infant expiratory crackle in comparison with various wavelets such as [1]. Du,morlet , Mexican Hat,Daubechies and Paul. It
 
Volume 2, Issue 9, September 
 – 
 2017 International Journal of Innovative Science and Research Technology ISSN No: - 2456
 – 
 2165 IJISRT17SP137 www.ijisrt.com 285 can be observed that crackle waveform has some similarity to the basic shape of the various wavelets. Figure 1: MFCC Plot of Pneumonia Signal. Figure 2: Side-By-Side Comparison of (A) Example Infant Expiratory Crackle. With Various Wavelets: (B) Morlet , (C) Du, (D) Daubechies3, (E) Mexican Hat, (F) Paul. Wavelet feature of cough sound can be extracted is given by The process will applied for slant wavelet transform following computation is used to calculation of CWT.[1] Computation of CWT:
 
Let
 x
denote an RMS normalized cough sample.
 
Apply CWT on scales. Let
ci
represent wavelet representation of
 x
on the
i
th scale, where
i
= 1
 ,
2
 ,
3
 ,
etc.
 
Segment each
ci
to equal non overlapping sub segments and calculate the energy concentration by sum of absoluteValues in each segment,
cij
, where
 j
= 1,2,3, . . . ,etc Eachcough sample,
c
i
 
For each
ci
, calculate the slopes for each
cij
along the timeAxis. For the first segment, it is the ratio of
cij
:
ci
(
 j
+1) .
 
For segments 2
 – 
11, it is the ratio of
ci
(
 j
1) :
ci
(
 j
+1). For The last segment, it is the ratio of
ci
(
 j
1) :
cij
.
 
Repeat for each ci for all cough samples
V.
 
EXPERIMENTAL SETUP
The cough sound can be collected from Datta Hospital Sangamner.most of the patients showing symptom of  pneumonia. The recording setup contains of high reliability recordings from one bedside microphone having the model  NT1 RODE. Software NUEND04 used for recording purpose. The distance between the microphone and subject is near about 1 foot. distance may be vary depend upon movement of the subject .we keep the sampling rate 48Khz sampling/s and 16 bit resolution to obtain the best quality of sound. File format for speech format is .wav.the total recording sample 300 collected from this study and split into the training and testing data set.
VI.
 
SYSTEM ARCHITECTURE
 A.
 
Choice of Scale
The selection of scale used for direct conversion of its 2 cycle duration to frequency. Selection of scale in wavelets is like that of window sizes in short time Fourier transform which determine the frequency resolution of cough sounds which is directly affects the shapes of the output of the signal. Figure 3: Block Diagram of Proposed System.
0 50 100 150 200 250 300 350 400 450-7-6.5-6-5.5-5-4.5

Reward Your Curiosity

Everything you want to read.
Anytime. Anywhere. Any device.
No Commitment. Cancel anytime.
576648e32a3d8b82ca71961b7a986505