e-issn 2455 1392 Volume 2 Issue 4, April 2016 pp. 271-275 Scientific Journal Impact Factor : 3.468 http://www.ijcter.com Monitoring Cardiac Stress Using Features Extracted From S1 Heart Sounds Biju V. G. 1, Anith Mohan 2, Akhilmon Sebastian 3, Nebu George 4,Anjana Rose 5, Mary Liya K A 6 1,2,3,4,5,6 Department of ECE, CEM Abstract- Acoustic heart sounds are known to provide a powerful tool for providing cardiac activity. Heart sound contains significant information on the mechanical activity of heart. This paper presents a method for the analysis of cardiac monitoring. The proposed study involves analysis of certain morphological features of the acoustic signals that are associated with physiological changes of heart with a trained network. The study shows that the proposed features vary significantly during cardiac stress. Keywords- S1 heart sounds, signal transformations, feature extraction I. INTRODUCTION Cardiovascular diseases are one of the leading causes of death worldwide. The use of remote monitoring devices in cardiology has been seen to significantly improve efficiency of healthcare services. Remote monitoring technique reduces cost and improves quality of service. The possible detection of a problem before it develops into a critical condition has great benefits and this could be achieved with just a regular examination done at home[5]. Most devices for remote cardiac monitoring are based on ECG. However, it does not provide detailed information on the state of heart valves and other important factors. ECG monitoring alone is not adequate to detect cardiac irregularities in real time. Specifically, conventional visual ECG monitoring for the detection of cardiac variations. Another method is to use vibro-acoustic heart signals. Analysis of vibro-acoustic heart signals is non-invasive, reliable and easy to perform. Heart sounds carry valuable information. Heart sound analysis involves various processes such as feature extraction and segmentation. Discriminative features are used for obtaining high accuracy[6]. Heart sounds are generated by closure of valves and flow of blood through it. The two major audible sounds in a normal cardiac cycle are the first and second heart sounds, S1 and S2. S1 occurs at the onset of the ventricular contraction during the closure of the mitral and tricuspid valves[1]. S2 is heard at the end of the ventricular systole, during the closure of the aortic and pulmonary valves. In this work the audio recordings of HSs from several patients at BL(baseline) activity and during cardiac stress are analysed. Here candidate features are extracted to distinguish the BL state from cardiac stress. These features includes mean, variance, kurtosis and skewness. The extracted features are applied to train the artificial neural network. An artificial neural network is an interconnected group of nodes. The trained features are then compared with new input features and cardiac stress is analyzed[4]. II. PROPOSED METHOD Heart sounds (HS) are generated by blood flow and the closure of valves. The HS signal of a healthy heart is composed of two distinct components S1 and S2[4]. For each patient two sets of S1 @IJCTER-2016, All rights Reserved 271
HSs(heart sounds) are recorded, a BL(base line) set which contains HSs recorded under anesthesia before the beginning of surgery and a Monitoring State set (MS). And BL is used to compare the features of heart sound signals. For each set the features like moment, mean, skewness and kurtosis are calculated. These features are used to compare previously recorded base line and monitoring state. Fig 1: Computational framework Noise removal: In this phase, removal of various types of noise which can distort the features of basic HS components is done. Various methods have been used to separate HS signals from lung sounds, other background noise and friction between the recording device and skin[3]. The input to this stage is a HS signal. The acquired vibro-acoustic signal is digitally filtered by a chebyshev type1 bandpass filter and it has a passband between 20-75 Hz. Signal transformation: The input to signal transformation stage are the filtered HS signal and output are these sets after transformation to a signal. In this work we examine three types of signal representation: wavelet transform, hilbert transform and fourier transform. Wavelets allow complex information such as music, speech, images and patterns to be decomposed into elementary forms at different positions and scales and subsequently reconstructed with high precision. Wavelets as a family of functions constructed from translations and dilations of a single function called the "mother wavelet" ψ(t). They are defined by: ( ) ( ) In signal processing, the Hilbert transform is a linear operator that takes a function, u(t), and produces a function, H(u)(t), with the same domain. ( )( ) ( ) ( ) The envelope of the heart sound signal is detected using Shannon s energy. Shannon energy is calculated using a windowing technique. The signal is windowed using hanning window which helps to smooth the edges and reduce the effect of large signal in adjacent time segments. Thus it helps to separate closely spaced subcomponents of S1 and S2. @IJCTER-2016, All rights Reserved 272
Segmentation: Segmentation of HSs into cardiac cycles involves partitioning into s1 and s2 components. A threshold is set on the output signal to detect S1 and S2 events [4]. Cardiac feature extraction and comparison: Features like mean, moment, skewness and kurtosis are extracted from the S1 HSs during baseline (BL) activity and during stress, and these features are compared [2]. Comparison is done by the use of neural network. An artificial neural network is an interconnected group of nodes. This neural network is trained using a set of database. Here classified the database into four categories; two normal sets and two abnormal sets. Then applied some other heart signals as input to check whether it belong to which category. The output from the neural network says the nature of the input given by comparing it with the trained database set. The system is noninvasive and portable and therefore it could also be used for home monitoring. III. RESULTS AND DISCUSSIONS Fig 2: Original Signal Fig 3: Filtered Signal @IJCTER-2016, All rights Reserved 273
Fig 4: S1 S2 segmentation Fig 5: S1 signal The figures represent the heart signal variations of one patient. Fig 2 shows unfiltered input heart sound signal. The input signal is filtered using Fig 3 shows the filtered output of original heart signal by chebyshev1 filter. Fig 5 represents S1 heart sounds separated from above Fig 4. From extracted S1 sounds, features like mean, skewness and kurtosis are calculated. From the input heart sound signal, we removed the noise by the use of chebyshev filter. The order of the filter is found and a filter of such an order is implemented. From the filtered signal found out the S1-S2 signal. By finding the odd and even sequence of that signal, it is possible to classify it into S1 and S2 heart signals separately. S1 signal is used for further analysis. For analysis of heart signal, some features are extracted from it. IV. CONCLUSION The work applied on two datasets of variable acoustic heart signals was able to accurately predict the physiological condition of the patient. On the first dataset the normal heart sounds were estimated @IJCTER-2016, All rights Reserved 274
while in the second dataset the level of stress is predicted. Hence this proposed method of analyzing heart sounds for monitoring cardiac stress is more efficient and accurate than the conventional method of using ECG. Thus provides a new technology for detection and diagnosis of mechanical disorders caused by cardiovascular diseases. REFERENCE [1] S. Hoeks et al., Cardiovascular risk assessment of the diabetic patient undergoing major noncardiac surgery, Best Practice Res. Clin. Endocrinol. Metabolism, vol. 23, no. 3, pp. 361 373, 2009. [2] A. Biagini et al., Unreliability of conventional visual electrocardiographic monitoring for detection of transient st segment changes in a coronary care unit, Eur. Heart J., vol. 5, no. 10, pp. 784 791, 1984. [3] Herzig, J., Bickel, A., Eitan, A. and Intrator, N., 2015. Monitoring Cardiac Stress Using Features Extracted From S1 Heart Sounds. Biomedical Engineering, IEEE Transactions on, 62(4), pp.1169-1178. [4] A. Biagini et al., Unreliability of conventional visual electrocardiographic monitoring for detection of transient st segment changes in a coronary care unit, Eur. Heart J., vol. 5, no. 10, pp. 784 791, 1984. [5] S. Kofman et al., Discovery of multiple level heart-sound morphological variability resulting from changes in physiological states, Biomed. Signal Process. Control, vol. 7, no. 4, pp. 315 324, 2012. [6] Z. Dokur and T. lmez, Heart sound classification using wavelet transform and incremental self-organizing map, Digital Signal Process., vol. 18, no. 6, pp. 951 959, 2008. [7] Z. Yan et al., The moment segmentation analysis of heart sound pattern, Comput. Methods Prog. Biomed., vol. 98, no. 2, pp. 140 150, 2010. [8] C. Ahlstrom et al., A method for accurate localization of the first heart sound and possible applications, Physiol. Meas., vol. 29, no. 3, pp. 417 428, 2008. [9] D. Gill et al., Detection and identification of heart sounds using homomorphic envelogram and self-organizing probabilistic model, in Proc. Comput. Cardiol., 2005, pp. 957 960. [10] C. N. Gupta et al., Neural network classification of homomorphic segmented heart sounds, Appl. Soft Comput., vol. 7, no. 1, pp. 286 297, 2007. [11] T. lmez and Z. Dokur, Classification of heart sounds using an artificial neural network, Pattern Recog. Lett., vol. 24, no. 1, pp. 617 629, 2003. [12] S. Choi and Z. Jiang, Comparison of envelope extraction algorithms for cardiac sound signal segmentation, Expert Syst. Appl., vol. 34, no. 2, pp. 1056 1069, 2008. [13] A. Moukadem et al., Phonocardiogram signal processing module for auto-diagnosis and telemedicine applications, 2012, [Online]. Available: http://www.intechopen.com/books/authors/ehealth-andremote- monitoring/phonocardi ogram-signal-processing-module-forauto- diagnosis-and-telemedicine-applications. [14] G. Livanos et al., Heart sound analysis using the S transform, in Proc. Comput. Cardiol., 2000, pp. 587 590. @IJCTER-2016, All rights Reserved 275