Classification of Heart Sounds Associated With Murmur For Diagnosis of Cardiac Valve Disorders


 Background: Now a day, cardiovascular diseases have been a major cause of death in the world. The heart sound is still the primary tool used for screening and diagnosing many pathological conditions of the human heart. The abnormality in the heart sounds starts appearing much earlier than the symptoms of the disease. In this study, the Phonocardiography signal has been studied and classified into three classes, namely normal signal, murmur signal and extra sound signal. A total of 15 features from different domains have been extracted and then reduced to 7 features. The features have been selected on the basis of correlation based feature selection technique. The selected features are used to classify the signal into the predefined classes using multi- class SVM classifier. The performance of the proposed denoising algorithm is evaluated using the signal to noise ratio, percentage root means square difference, and root mean square error. For this work a publically available database for researchers, Partnership Among South Carolina Academic Libraries (PASCAL) and MATLAB 2018a was used to develop the proposed algorithm.Results: Our experimental result shows that the 4th level of decomposition for the Db10 wavelets shows the highest SNR values when using the soft and hard thresholding. The overall accuracy, Sensitivity and Specificity of the developed algorithm is 97.96%, 97.92 % and of 98.0% respectively.Conclusion: even if the proposed algorithm is useful for murmur detection mainly valve-related diseases and the efficiency of the proposed study is increased, future work will intend to generalize the algorithm by using hybrid classifiers on a larger dataset. Since all experiments used the PASCAL datasets, additional experiments will be needed using new datasets to be implemented using the latest mobile phones which can work as an electronic stethoscope or phonocardiogram. In addition, the case of continuous murmur and types of murmur has been included for classification in further studies.


31
Heart disease is the main health problem and a primary cause of fatality all over the world. 32 Phonocardiography, tracing of sounds produced by digital stethoscope, is a tool that leads to valuable 33 PCG information about the heart function and can be a great tool to identify abnormalities and heart 34 disease early. Cardiovascular disorders (CVD) are the number one causes of death globally and more 35 people die annually from CVDs than from any other causes (1). A lot of studies show the proportion of 36 deaths due to non-communicable diseases under the age of 70 years while Cardiovascular diseases 37 assume the largest proportion of deaths among the non-communicable diseases deaths (51.45%), 38 followed by cancers and chronic respiratory diseases. All broad indications derived from a range of 39 developing countries indicate an increasing burden imposed by cardiovascular diseases (1)(2). 40 Cardiovascular disorders are broad terms that can affect both vasculature and the heart muscle itself. In 41 auscultation technique, a stethoscope is used for heart sound analysis to diagnose the condition of 42 human heart generated by muscle contractions and closure of the heart valves which produces 43 vibrations audible as sounds and murmurs, which can be analyzed by qualified cardiologists (3). The 44 existence of murmur in PCG recording is often related to heart valve diseases. Heart diseases include; 45 heart failure, coronary artery disease, hypertension, cardiomyopathy, valve defects, and arrhythmia. 46 The current study is concerned only with heart valve defects. There are two general types of cardiac 47 valve defects: stenosis and insufficiency. Valvular stenosis results from a narrowing of the valve orifice 48 that is usually caused by a thickening and increased rigidity of the valve leaflets, often accompanied by 49 calcification. When this occurs, the valve does not open completely as blood flows across it. Valvular 50 insufficiency results from the valve leaflets not completely sealing when the valve is closed so that 51 regurgitation of blood occurs (backward flow of blood) into the proximal chamber(4). 52 The heart sounds is still the primary tool for screening and diagnosing many pathological conditions of 53 the human heart, which is compound sound of mechanical vibration, and involves different parts of the 54 heart. Conventional auscultation using acoustic stethoscope requires extensive training and experience 55 of the physician for proper diagnosis. Moreover, the storage of records for follow-ups and future 56 references is also not possible with conventional auscultation (5). This is the driving force for this study 57 in order to move towards automatic auscultation using electronic stethoscopes. In the current study 58 PCG will be used for heart condition monitoring which finds its roots in auscultation. There is difficulty in 59 performing conventional heart sound diagnosis. The main issues are difficulty of obtaining high quality 60 signals, the differences in hearing sensitivity of each person and the vast amount of experience to 61 master heart auscultation skills (3). 62 Murmurs are caused by blood turbulence which is capable of producing a sound that can be heard using 63 a stethoscope. The murmurs can be termed as indicators of various heart problems (6). The problem 64 causing murmurs could be congenital or developed with time. As heart sounds and murmurs have very 65 less overlap with human audibility range, the minute details that can be missed during auscultation can 66 be best viewed and taken care of with the help of PCG. 67 Many researchers have been proposed different methods on how heart diseases can be diagnosed. So 68 far, an intelligent algorithm based on PCG signal analysis (7), a new analytical technique called Digital 69 Subtraction Phonocardiography (8), which is based on the principle that the murmurs are random in 70 nature, measuring entropy to analyze heart sounds (9) and a new feature called mean12 was proposed, 71 which is the maximum of the mean in the systolic and diastolic region to classify signals into two classes 72 i.e. normal and murmur signal (10). The aim of this work is to develop a system for classification of 73 pathological heart sounds associated with murmur for diagnosis of cardiac valve disorder by using DWT 74 and multi-class support vector machine learning algorithm. Therefore, PCG signal is investigated in time, 75 frequency and statistical domain. Additional features were also introduced to increase the efficiency and 76 accuracy of the proposed method. 77

78
In this study, the PCG signals were studied and classified into three classes, namely normal 79 signal, murmur signal and extra sound signal. Many features in time, frequency and statistical 80 domains have been extracted and the best features were selected for the classification using 81 Multi SVM which is illustrated in Tables 2 and 3. Two new features mid-frequency and average 82 frequency were introduced in this study for the classification of the PCG signals. The study also 83 presents the application of the wavelet transform method to PCG signal noise elimination which is 84 examined at different levels and the Db10 wavelets at the 4th level of decomposition offer the 85 maximum SNR and minimum RMSE for HS. In this work classification method is proposed to 86 separate normal and abnormal heart sound signals having murmurs without getting into the 87 cumbrous process of segmenting fundamental heart sounds using ECG gating. Thus it will have a 88 good potential to help researchers who need to study heart diseases identification based on heart 89 sounds (classifying normal heart sounds from pathological murmur) and also applicable for the 90 development of portable devices.

91
The accuracy of the presented algorithms can be further increased by incorporating Artificial Intelligence 92 techniques or other hybrid classifiers on a larger dataset. The case of continuous murmurs and its types 93 are not included in the study. So it can be included for classification in further studies. 94

95
The work performed in preprocessing is to determine the most suitable parameters for a wavelet 96 algorithm to denoise heart sound signals with excellent ability to inform physicians about heart related 97 problems. This is by adding white noise to the original signals and applying different types of wavelet 98 thresholding to remove the noise from the PCG signals, with different thresholding rules (Rigrsure, 99 Sqtwolog, Heursure, and Mini-max) to analyze the resulting denoising performance of PCG signal. After 100 applying a threshold at each level of the original signal, the effects of noise on PCG signals were 101 removed. Finally, the denoised signal was reconstructed using IDWT. The algorithm was tested using the most widely used wavelet families, i.e., Daubechies wavelet family, 111 Symlets wavelet family, Coiflets wavelet family and discrete Meyer wavelet family, The tested PCG 112 signals were contaminated by white noise added at SNR = 5 dB as an initial value to test the 113 performance of the proposed technique for noise elimination. Figure 1 shows the wavelet coefficients of 114 the denoised signal, whereas Figures 2 and 3 show the effect of the Sym6 and Db10 wavelets on 115 denoising the normal PCG signal using the 4 th level of decomposition. Figure 4 shows a histogram 116 comparing the SNR values obtained when using the different wavelet families with soft and hard 117 thresholding. To study the effect of the two thresholding types Table 1  domain only. Therefore, Figure 4 presents spectrograms for the noisy and denoised PCG signals to show 141 the clarity of the heart sound components obtained after applying the proposed denoising algorithm. In 142 the denoised PCG signal spectrogram, the heart sounds are clear. The features which are extracted in the feature extraction phase are then reduced to a few features 150 which are further used for classification. This is done in order to reduce the dimensionality, redundancy 151 and computational load. The features that have been reduced using CFS and those selected features 152 with higher CFS values are shown in Table 3. 153

207
The methodology used in this study to classify various heart sounds into predefined classes consists of 208 five stages i.e. Signal acquisition, preprocessing, feature extraction, feature reduction, and classification 209 as shown in Figure 5. 210

Datasets 211
The PCG signal acquisition can be done by electronic stethoscopes which respond to the sound waves 212 identically to the conventional acoustic stethoscope with the changes in electric field replacing the 213 changes in air pressure. For this study an electronic database of PCG signals was taken from PASCAL 214 (11). The dataset used was taken from a clinical trial in hospitals using a digital stethoscope from adults. 215 In this study, a dataset recorded from PCG having 300 signals was used out of which 150 are normal 216 signals, 100 are murmur signals and 50 are extra sounds. 217

Wavelet-based Preprocessing of PCG signals 218
Heart sound signal is a typical biomedical signal, which is random and has strong background noise. In 219 the process of collecting heart sound signals, it is vulnerable to external acoustic signals and electrical 220 noise interference; particularly , friction caused by subjects breathing or body movement(12). The main 221 idea of the wavelet denoising algorithm is to obtain the essential components of the signal from the 222 noisy one, then threshold the small coefficients considering them to be pure noise. In this research, four The discrete wavelet transform was used to extract characteristics from a signal on various scales 265 proceeding with successive high pass and low pass filtering. The wavelet coefficients are the successive 266 continuation of the approximation and detail coefficients. The basic feature extraction procedure 267 consists of decomposing the signal using DWT into N levels using filtering and decimation to obtain the 268 approximation and detailed coefficients and extracting the features from the DWT coefficients as shown 269 in Table 6. 270 The various steps involved in the feature extraction algorithm are summarized as follows: 271 1. The HS signal decomposes into four detail subbands using discrete wavelet transform. The 6. The features are computed either by using syntax or by implementing the formula. they are mean, 282 variance, standard deviation, kurtosis, skewness, root mean square, total harmonic distortion, 283 bandwidth, dynamic range, maximum amplitude, cepstrum peak amplitude, power, average 284 frequency, maximum frequency, and mid frequency. 285  winner takes all to predict the right answers, but only the correct label will have a positive score. In this 342 study, this algorithm is selected because it is easy to learn and use any binary classifier learning 343 algorithm. 344