Spectral and SNR improvement analysis of normal and abnormal heart sound signals using different windows

Accepted Manuscript Spectral and SNR improvement analysis of normal and abnormal heart sound signals using different windows G. Rajkumar, R. Jayabharathy, K. Narasimhan, N. Raju, M. Easwaran, V. Elamaran, Gustavo Ramirez-gonzalez, Marlon Burbano-fernandez

PII: DOI: Reference:

S0167-739X(18)31813-2 https://doi.org/10.1016/j.future.2018.09.047 FUTURE 4482

To appear in:

Future Generation Computer Systems

Received date : 29 July 2018 Revised date : 5 September 2018 Accepted date : 17 September 2018 Please cite this article as: G. Rajkumar, et al., Spectral and SNR improvement analysis of normal and abnormal heart sound signals using different windows, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.09.047 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Spectral and SNR Improvement Analysis of Normal and Abnormal Heart Sound Signals using different Windows 1 G. Rajkumar1, R. Jayabharathy1, K. Narasimhan *, N. Raju1, M. Easwaran21, V. Elamaran1, 2 Gustavo Ramirez-gonzalez and Marlon Burbano-fernandez 1

Department of ECE, School of EEE, SASTRA Deemed University, Thanjavur, Tamil Nadu, India. 2 Department of Telematics, University of Cauca, Colombia. *Corresponding author: [email protected]

 

ABSTRACT The primary function of a coronary care unit is to monitor the rhythm of cardiac patients. The cardiac abnormality of the patients should be identified during the early stage. The heart sound recoding is the phonocardiogram (PCG) signal. Stethoscopes are commonly used to hear the heart sound for diagnosis. In common, the heart murmurs can be caused by a high blood pressure, heart attack, pregnancy, rheumatic fever, anemia or thyrotoxicosis. This study analyzes PCG signals spectrally with different windows with signal-to-noise ratio (SNR) computations. The two PCG signals of normal and abnormal subjects in each were analyzed in this study. The normal subject produce clear “lub” and “dub” sounds where the abnormal subject produce a kind of whistling or swishing sound in middle which may be due to the problem of septal defect in a heart. The spectral leakage reduction by different windows helps to increase the SNR value. Simulation results of SNR with different windows are obtained using Matlab R2016b tool. Keywords: Heart sound, PCG, spectral leakage, signal-to-noise ratio, window. 1. INTRODUCTION The heart is an important organ with an average size about 12 cm long (base-to-apex) and 8 cm wide. Its principal functions are to pump the deoxygenated blood to the lungs in which the exchange of carbon dioxide-oxygen takes place, and to pump the oxygenated blood to the whole body. The right and left sides are the two functional sides of a heart. They are separated by the septum. They are further divided into chambers. The right-sided chambers contain the right atrium and the right ventricle; the left-sided chambers contain the left atrium and the left ventricle. Figure 1 shows the cross section of a human heart [1,2]. The chambers are separated by unidirectional valves which control the blood flow too. The right atrium and right ventricle are separated by the right atrioventricular (AV) valve; the left atrium and left ventricle are separated by the left AV. The right ventricle and pulmonary artery are separated by the semilunar pulmonic valve; the left ventricle and aorta are separated by semilunar aortic valve. The vibrations of from bumpy blood flow, heart’s walls and valves are heard as heart sounds through a stethoscope. During the cardiac cycle, the heart’s walls and valves are flexible as they move in response to pressure [1,2].

Fig. 1: Human heart cross section In human body, to diagnose the clinical auscultation, the heart sound (HS) is an important physiological signal [3–7]. Two basic heart sounds (S1 and S2) are produced from a normally functioning heart during a a cardiac cycle. These sounds are caused due to the blood acceleration or deceleration in the heart’s chambers. S1 produces the “lub” sound and S2 produces the “dub” sound, and hence “lub-dub”, “lub-dub”, sounds occur in a heart. Even an ordinary person can hear and identify or distinguish the different heart beats such as normal heartbeat, fast heartbeat, slow heartbeat and irregular heartbeat [8–11]. Heart sound or phonocardiogram (PCG) signals do have high similarity with the electrocardiogram (ECG) signals. In ECG, during the cardiac cycle, the activation of the atria belongs to P wave; the activation of the ventricles belongs to QRS complex; the recovery wave belongs to T wave [12–16]. The ventricular contraction is the reason for the first heart sound “S1”. This S1 sound occurs at the same time as the QRS complex in the ECG signal with a low frequency band about 10 – 120 Hz. The second heart sound “S2” is due to the closing scenario of the pulmonary and aortic valves, which occurs during the end of the T wave in the ECG signal. The S2 sound frequency is normally higher than that of the first sound with a low frequency band about 10 – 200 Hz [17]. The normal heart rate is required for healthy people. Abnormal heart rates such as, above the normal is referred to as “tachycardia”; the below normal one is referred to as “bradycardia” [17]. For example, new born babies heart rate should be around 130 beats per minute; Childs heart rate should be in the range between 80 to 100 beats per minute; adults should have in the range between 70 to 90 beats per minute; athletes should have in the range between 50 to 70 beats per minute [19].

The heart sounds are produced due to the blood acceleration or deceleration in the heart’s chambers. Heart sound or phonocardiogram (PCG) signals do have high similarity with the electrocardiogram (ECG) signals. In ECG, during the cardiac cycle, the activation of the atria belongs to P wave; the activation of the ventricles belongs to QRS complex; the recovery wave belongs to T wave [12–16]. This paper is organized as follows. Section 2 contains the equations of all windows used on normal and abnormal heart sound signals for the spectral analysis and Section 3 includes simulation results. The conclusions are finally described in Section 4. 2. MATERIALS AND METHODS The different windows are described here. The spectral and SNR analysis on normal and abnormal heart sound signals are done using these windows and the results are discussed for a comparative study in the next section. 2.1 Rectangular window

In the spectral analysis, actually, the sampled portion of the signal is taken without any change. This window has the stopband attenuation about 21 dB irrespective of the window size. All windows have the property that the transition width is increased with more window points not the stopband attenuation [19]. Other windows have better stopband attenuation than the rectuangular window. 2.1 Rectangular (Dirichlet) window

In the spectral analysis, actually, the sampled portion of the signal is taken without any change. This window has the stopband attenuation about 21 dB irrespective of the window size. All windows have the property that the transition width is increased with more window points not the stopband attenuation which is illustrated in Figure 2. For example, as in Figure 2, the peak of the first side lobe of rectangular window of length 11 and 51 are around –13 dB. Other windows have better stopband attenuation than the rectangular window. The equation for the rectangular window is [19,20]:

Wrect (n) 1, 0  n  N  1

(1)

Fig. 2: Spectral characteristics of a rectangular window of length: (a) 11 and (b) 51 2.2 Hanning window

The von Hann, cosine bell and raised cosine are the other names available for the Hanning window. This window provides better stopband atteneuation (50 dB) but with more transition width, approximately three times higher than the rectangular window. This window is expressed as [19,20],

 2 n  Whann (n)  0.5  0.5 cos  , 0  n  N  1  N 1

(2)

2.3 Hamming window

This window provides better stopband attenuation, i.e., 10 dB higher than the hanning window but at the cost of wider transition width. The equation of Hamming window is [19,20]:  2 n  Whamm (n)  0.5  0.46 cos  , 0  n  N  1  N 1 2.4 Blackman window

(3)

The tradeoff between the transition width and a stopband attenuation can be achieved using Blackman window. This window provides 74 dB stopband attenuation but at the cost of wider transition width, i.e., six times higher than the rectangular. This window function is expressed as [19,20],

N 1  2 n   2 n  Wblackman (n)  0.42  0.5 cos  , n    0.08 cos  2  N 1  N 1

(4)

2.5 Bartlett window

The peak of the first side lobe of this window is around 26 dB. Since this is a triangle, straight lines are used for tapering. Conceptually, the convolution of two rectangular windows become a Bartlett window [19–21]. sdsd Based on zeros inclusion at the end of tapering, this window function can be a similar one to a triangular window. The pattern which includes zeros at the edges is referred to as “triangular window”; otherwise remains “Bartlett window”. The Bartlett window is expressed as [21, 22, 23, 24, 25], Wbartlett (n) 1 

|n| N 1 , 0n N /2 2

(5)

2.6 Bohman window

The Bohman window of length – 101 and its Amplitude spectrum are shown in Figure 3 with 46 dB – the peak of the first side lobe. The expression for this window is expressed as [26,27],

|n|  N   |n|  1  |n|  Wbohman (n)  1   cos     sin   , 0  n  2  N /2  N /2   N /2

Fig. 3: Bohman window of length – 101 and its Amplitude spectrum 2.7 Parabolic and Riemaan windows

(6)

The parabolic window is expressed as [27], N  n 2 W parabolic (n) 1    N /2   and the equation of the Riemann window is [27]:

2

   , 0  n  N 1   

2 n N , 0n N WRiemann (n)  2 n 2 N

(7)

sin

(8)

2.8 Parzen and Tukey windows

The Parzen window is expressed as [27], 2  n  N   n    1 6 1  , 0   N      N / 2  4    N / 2 W ( n)   3   n  N N  2 1   n  ,  4 2   N / 2 and the equation of the Cosine Taper (Tukey) window is [27]:

N   1, 0 n   2   W (n)    0.5 1  cos  n   ( N / 2)  ,  N  n  N   (1   )(n / 2)     2 2      

(9)

(10)

2.9 Hann-Poisson and Gaussian windows

The Hann-Poisson window expressed as [27],

 n   n  N  , 0  n  W (n)  0.5 1  cos    exp     N /2 2  N / 2   

(11)

and the equation of the Gaussian window is [27, 28]: 2  1 n   N W (n)  exp     , 0  n  2  2  N / 2  

3. EXPERIMENTAL RESULTS AND DISCUSSION

(12)

The recording of the heart sounds is referred to as phonocardiogram (PCG). The frequency content and the maximum intensities of the S1, S2 and murmur sounds, the intensity patterns of murmurs, and timing of S1, S2 and murmurs are the important features of the PCG. 3.1 PCG Signal Dataset

In this study, four patient PCG signals are taken in which two are normal (“pcgn1.dat” and “pcgn2.dat”) and the other two are abnormal (“pcgab1.dat” and “pcgab2.dat”) [17].That is, pcgn1 and pcgn2 have the samples from normal patients; pcgab1 and pcgab2 have the samples from abnormal patients. Here, the abnormal patients have the problem of ventricular septal defect. The occurrence of a hole in between the two ventricles is referred to as the septal effect, in which the blood leak happens from the left ventricle to the right ventricle during the systole period. This results systolic murmurs in the PCG signal. Figure 4a and 4b illustrate the PCG signal, the portion of the PCG signal (9462 samples) and the Hanning windowed portion of the PCG signal of the normal subject 1 and 2, respectively. Figure 5a and 5b illustrate the PCG signal, the portion of the PCG signal (9462 samples) and the Hanning windowed portion of the PCG signal of the abnormal subject 1 and 2, respectively. It is apparent that the abnormal subjects have more murmur sounds than the normal subjects.

Fig. 4: The PCG signal, the portion of the PCG signal and Hanning windowed portion of the PCG of (a) Normal subject 1; (b) Normal subject 2

Fig. 5: The PCG signal, the portion of the PCG signal and Hanning windowed portion of the PCG of (a) Abnormal subject 1; (b) Normal subject 2 3.2 Signal-to-Noise Ratio Results

Figure 6a and 6b show the SNR results using different windows with normal and abnormal subjects respectively. Simulation results reveal that Tukey and triangular windows obtain better SNR results as 182.3033 dB and 182.1685 dB for the normal subject 1 respectively. The parabolic and Bartlett windows provide better SNR results as 176.4327 dB and 176.1895 dB respectively. The Tukey and Parzen windows provide better SNR results as 178.2227 dB and 175.8554 dB respectively. The parabolic and Riemann windows obtain better SNR results as 175.2689 dB and 162.9955 dB respectively.

SNR (dB)

SNR with different windows 200 180 160 140 120 100 80 60 40 20 0

Fig. 5: SNR results using different windows with (a) Normal subjects; (b) Abnormal subjects. 4. CONCLUSION

The detailed spectral and SNR results analysis is done in this study using different windows. The SNR improvement for a particular signal will help to interpret the data in a better way. That is, the spectral leakage issues are reduced using different windows effectively. The more spectral leakage would misinterpret the data highly. And hence this study becomes a primary important for the clinicians to diagnose the patient appropriately. This work can further be extended to morphological analysis and pattern classification of PCG signals.

REFERENCES

1. S.Ari, K.Sensharma and G.Saha, 2008. DSP implementation of a heart valve disorder detection system from a phonocardiogram signal. Journal of Medical Engineering & Technology, 32: 122–132. 2. Jessica Shank Coviello, 2013. Aucultation Skills: Breath and Heart Sounds. Lippincott Williams and Wikins,5th Edition. 3. Shi-Wen Deng and Ji-Qing Han, 2018. Adaptive overlapping-group sparse denoising for heart sound signals. Biomedical Signal Processing and Control, 40: 49–57. 4. V.N.Varghees and K.Ramachandran, 2014. A novel heart sound activity detection framework for automated heart sound analysis. Biomedical Signal Processing and Control, 13: 174–188. 5. A.Srinivasan, S.Krishnamurthi and A.Mathialagan, 2000. A System for AR Modeling Study of Cardiac Sound Signal. IETE Journal of Research, 46: 87–91. 6. Mazin Z.Othman and Asmaa N.Khaleel, 2017. Phonocardiogram Signal Analysis for Murmur Diagnosing using Shannon Energy Envelope and Sequenced DWT Decomposition. Journal of Engineering Science and Technology, 12: 2393–2402. 7. Sunjung Kim and Dosik Hwang, 2016. Murmur-adaptive compression technique for phonocardiogram signals. Electronics Leters, 52: 183–184. 8. Wenjie Zhang, Jiqing Han and Shiwen Deng, 2017. Heart sound classification based on scaled spectrogram and partial least squares regression. Biomedical Signal Processing and Control, 32: 20–28. 9. Yutaka Kagaya, Masao Tabata, Yutaro Arata, Junichi Kameoka and Seiichi Ishii, 2017. Variation in effectiveness of a cardiac auscultation training class with a cardiology patient simulator among heart sounds and murmurs. Journal of Cardiology, 70: 192–198. 10. Ting Li, Hong Tang, Tianshuang Qiu and Yongwan Park, 2013. Medical Engineering & Physics, 35: 1831–1836. 11. M.Abo-Zahhad, Sabah M.Ahmed and Sherif N.Abbas, 2016. Biometrics from heart sounds: Evaluation of a new approach based on wavelet packet cepstral features using HSCT-11 database. Computers & Electrical Engineering, 53: 346–358. 12. Amir A.Sepehri, Arash Gharehbaghi, Thierry Dutoit, Armen Kocharian and A.Kirani, 2010. A novel method for pediatric heart sound segmentation without using the ECG. Computer Methods and Programs in Biomedicine, 99: 436–48. 13. Koksoon Phua, Jianfeng Chen, Tran Huy Dat and Louis Shue, 2008. Heart sound as a biometric. Pattern Recognition, 41: 906–919. 14. Nele Famaey, Korijn Defever, Paul Bielen, Willem Flameng, Jos Vander Sloten, Paul Sas and Bart Meuris, 2010. Acoustical analysis of mechanical heart valve sounds for early detection of malfunction. Medical Engineering & Physics, 32: 934–939. 15. Nandini Sengupta, Md Sahidullaah and Goutam Saha, 2016. Lung sound classification using cepstral-based statistical features. Computers in Biology and Medicine, 75: 118–129.

16. Ana Oliveira and Alda Marques, 2014. Respiratory sounds in healthy people: A systematic review. Respiratory Medicine, 108: 550–570. 17. Rangarajan M. Rangayyan, 2009. Biomedical Signal Analysis: A Case-Study Approach. Wiley. 18. Robert M.Kliegman, Larry A.Greenbaum and Patricia S.Lye, 2004. Practical Strategies in Pediatric Diagnosis and Therapy. Saunders, 2nd Edition. 19. D.C. Reddy, 2005. Biomedical Signal Processing. McGraw Hill Education. 20. Dale Grover, John R. Deller, 1998. Digital Signal Processing and the Microcontroller. Prentice Hall. 21. R. Naveena, D. Rabecka, G. Rajkumar and V. Elamaran. Understanding digital filters-from theory to practice using Matlab and Simulink. International Journal of Pharmacy and Technology, 7: 9923–9934. 22. M.T. Naseem, K.R.A. Britto, M.M. Jaber, M. Chandrasekar, V.S. Balaji, G. Rajkumar, K. Narasimhan and V. Elamaran, 2017. Preprocessing and signal processing techniques on genomic data sequences. Biomedical Research, 22: 10205 - 10209. 23. V. Elamaran, T. Nirmalraj, S. Radhakrishnan, V.S. Balaji and Har Narayan Upadhyay, 2017. Experimental comparison of moving average FIR filter design architectures – A hardware redundancy approach. Journal of Engineering Technology, 6(Special Issue): 329 – 339. 24. V. Elamaran, Har Narayan Upadhyay, N. Raju and K. Narasimhan, 2015. Real time audio denoising using FIR filters with FPGA implementation. International Journal of Pharmacy and Technology, 7: 9923 – 9934. 25. T. Anusuya, R. Pratheepa and V. Elamaran, 2015. Removal of harmonic distortion in audio signals using wavelet packet decomposition. International Journal of Applied Engineering Research, 10: 16787 – 16796. 26. John G. Proakis, Dimitris K. Manolakis, 2006. Digital Signal Processing: Principles, Alogirthms, and Applications. Pearson Education, 4th Edition. 27. Silvia Maria Alessio, 2016. Digital Signal Processing and Spectral Analysis for Scientists: Concepts and Applications. Springer. 28. A.D. Poularikas, 1998. Handbook for Formulas and Tables for Signal Processing. CRC Press.

G. Rajkumar received the B.E. degree in Electronics and Communication Engineering, the M.E. degree in  Applied Electronics, and the Ph.D degree in Wireless Adhoc Networks from Anna University, India. His  research  interests  include  Digital  Signal  Processing,  Signal  Processing  Algorithms  and  Architectures,  Wireless Sensor Networks. He has published more than 30 papers in reputed international journals and  conferences. He is currently working as Senior Assistant Professor in the Department of ECE, School of  EEE,  SASTRA  Deemed  University,  Thanjavur.  He  is  a  Life  Member  of  the  Indian  Society  of  Systems  for  Science and Engineering (ISSE).  R. Jayabharathy received the B.E. degree in Electronics and Communication Engineering from Madurai  Kamaraj University, the M.E. degree in Power Electronics from Bharathidasan University, and the Ph.D  degree in Wireless Communication from SASTRA University, India. Her research interests include Digital  Signal  Processors,  Network  Theory,  and  Wireless  Communication.  She  has  published  more  than  30  papers in reputed international journals and conferences. She is currently working as Assistant Professor  in the Department of ECE, School of EEE, SASTRA Deemed University, Thanjavur. She is a Life Member of  the Indian Society of Systems for Science and Engineering (ISSE).  K. Narasimhan received the M.Sc. degree with Electronics Specialization from Bharathidasan University,  M.Tech. in Non destructive Testing from Regional Engineering College, Trichy and the Ph.D degree from  SASTRA University in the field of medical image processing. His research interests include Digital Image  Processing,  Medical  Image  analysis,  Pattern  Recognition,  Digital  Signal  processing.  He  has  published  more  than  40  papers  in  reputed  international  journals  and  conferences.  He  is  currently  working  as  Senior  Assistant  Professor  in  the  Department  of  ECE,  School  of  EEE,  SASTRA  Deemed  University,  Thanjavur. He is a Life Member of the Indian Society of Systems for Science and Engineering (ISSE).  N. Raju received the M.E. degree in Applied Electronics from Anna University, and the Ph.D degree in  Speech  Processing  from  SASTRA  Deemed  University,  India.  His  main  research  interests  are  robotics,  embedded systems, VLSI design, speech processing, and machine learning. He has published more than  30  research  papers  in  reputed  international  journals  and  conferences.  He  is  currently  working  as  Assistant Professor in the Department of ECE, School of EEE, SASTRA Deemed University, Thanjavur. He  is a Life Member of the Indian Society of Systems for Science and Engineering (ISSE).  M.  Easwaran  received  the  B.E.  degree  in  Electronics  and  Communication  Engineering  and  the  M.E.  degree  in  Power  Electronics  and  Drives  from  Bharathidasan  University,  India.  Currently  he  is  pursuing  Ph.D. in the area of Electrical Solitons from SASTRA Deemed University, Thanjavur, India. His research  interests include Power Electronics, Electronic Circuits, and Electrical Solitions. He has published more  than 10 papers in reputed international journals and conferences. He is currently working as Assistant  Professor in the Department of ECE, School of EEE, SASTRA Deemed University, Thanjavur. He is a Life  Member of the Indian Society of Systems for Science and Engineering (ISSE).  V.  Elamaran  received  the  B.E.  degree  in  Electronics  and  Communication  Engineering  from  Madurai  Kamaraj  University,  and  the  M.E.  degree  in  Systems  Engineering  and  Operations  Research  from  Anna  University,  India.  Currently  he  is  pursuing  Ph.D.  in  the  area  of  Low  Power  VLSI  Design  from  SASTRA 

Deemed University, Thanjavur, India. His main research interests are signal, image and video processing,  digital VLSI design circuits, design for testability, and FPGA based systems. He has published more than  80  research  papers  in  reputed  international  journals  and  conferences.  He  is  currently  working  as  Assistant Professor in the Department of ECE, School of EEE, SASTRA Deemed University, Thanjavur. He  is a Life Member of the Indian Society for Technical Education (ISTE) and the Indian Society of Systems  for Science and Engineering (ISSE).      Gustavo Ramirez‐gonzalez  Gustavo  Ramírez González received the B.S. degree in  electronic  and telecommunications engineering  from the University of Cauca, Cauca, Colombia, in 2001, and has a M.S. degree in telematics engineering  from  the  same  university.  He  received  his  Ph.D.  degree  in  telematics  engineering  from  the  Carlos  III  University in Spain, in 2010. He is currently a professor and researcher at the Department of Telematics  at  the  University  of  Cauca.  He  has  participated  in  national  and  international  projects  in  Colombia  and  Spain.  His  research  interests  include  image  processing,  secure  communication,  machine  learning  and  IoT.  He  has  published  several  research  papers  in  reputed  journals  and  served  as  a  Guest  Editor  for  several Special Issues at many journals    Marlon Burbano‐fernandez  Marlon  Felipe  Burbano  He  is  Electronics  and  Telecommunications  Engineer  since  2010,  Master  in  Telematics Engineering since 2015 and PhD student in Telematics Engineering 2018 at Universidad del  Cauca  ‐  Colombia.  He  has  been  linked  to  research  projects  with  international  and  national  impact  associated with ICT. In his doctoral studies, his interests have turned to Internet of Objects, especially  wearable devices for capturing data of the movements of the human body, in order to obtain different  patterns to use algorithms for data analysis. 

G. Rajkumar:

R. Jayabharathy:

K. Narasimhan:

N. Raju:

M. Easwaran:

V. Elamaran:

Gustavo Ramirez-gonzalez:

Marlon Burbano-fernandez:

Highlights 

This study analyzes PCG signals spectrally with different windows with signal-to-noise ratio (SNR) computations.  

The two PCG signals of normal and abnormal subjects in each were analyzed in this study.  

The spectral leakage reduction by different windows helps to increase the SNR value.