Study of Heart Sound Analysis Instrument

Transcription

1 Study of Heart Sound Analysis Instrument Xiefeng Cheng 1, 2*, zhang Saobai 1, cheng jianbing 1, Chen Hong 1 1 College of Electronics science & Engineering, Nanjing University of Posts and Telecommunications, Nanjing, China 2 Key Lab of Broadband Wireless Communication and Sensor Network Technology ( Nanjing University of Posts and Telecommunications), Ministry of Education, China Accepted: 22 October 2012 A heart sound analysis instrument based on the LabVIEW is devised. There are five subsystems in this instrument: heart sound receiving subsystem, wavelet de-noise subsystem, time domain analysis subsystem, frequency domain analysis subsystem and heart sound signal generation subsystem. The instrument is developed in PC, by using homemade wireless heart sounds receiving device and heart sounds signal receiving subsystem to get heart sound signals, and then using wavelet de-noising subsystem to remove the background noise, finally, the instrument uses time-domain analysis subsystem and frequency domain analysis sub-system to analyze the heart sounds signals. Heart sound signal generator subsystems can produce synthetic heart sound signals accord to the need of users to learn and use. The operating environment of this instrument proves it can collect vivid heart sound signal, reduce noise effectively, computer the characteristic values quickly and accurately, generate and play heart sound precisely. It can be used as an assistance to show, play, and analyze heart sound. Keywords: Heart Sound Analysis Instrument, Wireless Heart Sound Sensor, Time and Frequency Domain Analysis, Heart sounds signal generator INTRODUCTION With the development of microcomputer and software technology, a new generation of test instruments - virtual instrument appeared. Virtual instrument is able to share the hardware and software resources and establish various automatic test systems fast and conveniently. It is the development direction of instruments. Heart sounds signal collection and analysis has important significance in medical diagnosis, research and teaching. Various heart sounds analysis in time domain and frequency domain can help doctors make the correct diagnosis and assist researchers to do analysis research on all sorts of heart sounds. Heart sound signal generator can make scientific and study researchers avoid searching kinds of heart sounds, it can also help learners master various heart sound features fast in medical teaching. Therefore, it is meaningful to develop heart sound signal generators and multi-function treatment instruments based on virtual instrument (Jiang at el.,2006; Tanner at el., 2008). This instrument is developed in an ordinary PC and it uses homemade wireless heart sounds sensor, its software platform is based on the LabVIEW and Matlab. The instrument s system diagram is shown in figure 1. Heart sound analysis instrument has advantages for further analysis, research and teaching. *Corresponding author. chengxf@njupt.edu.cn; jnucxf@163.com. Tel: (Mobile) Figure 1 The diagram of heart sounds analyzer system based on the LabVIEW 1. WIRELESS HEART SOUND DETECTION DEVICE The heart sounds signals model can be expressed as follows (Cheng at el., 2012): 4 S t S t S t S t S t S t (1) = ( )+ 4 = i i=1 The heart sounds signals s(t) have characteristics such as: (1)The heart is a complex system with nonlinear and timevarying characters. It also has the characters of strong non-stationarity and randomness which caused by many uncertainties such as breath, cardiac-hemodynamics and the environment where people are. (2) Heart sound signals are similar, but they all have

2 nuances with each other. Frequency of the heart sound concentrate on the range of frequency from 0 ~ 600Hz. Frequency of the first heart sound is between 50 and 150 Hz while one of the second heart sound is between 50~200 Hz. There is a second peak in the range of 250~350 Hz. They have obvious change on amplitude while the other frequency range is tend to zero and there is the nuance between different signals. Thus, heart sound identity recognition need pick up the details of signal by special equipment for fully present the personality characteristics of heart sound signals. (3) The heart sound signal is periodic as the heartbeat is periodic. Although the waveform of each cycle have nuances and interference. We can still deal with heart sound signal by regarding it as repeated stable signal when we sound of heart identity recognition.. (4) The information of heart sound signal mainly focus on the first heart sound and the second heart sound, which are the dominating place of the sound of heart identification of parameters. However the third heart sound and forth heart sound is inconspicuous, so they are not underlined. Figure1 shows the characteristics said before by a heart sound signal. Where Figure 2 (a) a normal sound signal with sampling frequency of 11025Hz.Figure 2 (b) is the waveform divided by period. The repeated time is 0.93s. Figure 2(c) is the waveform and power spectrum of s1 and s2 in one period. hardware part mainly consists of computers and homemade wireless voice detection device Since the wireless peripherals with 2.4G frequency has advantages of low power, far distance and little interference, we adopt wireless communication technology with 2.4G frequency to solve the wireless communication problems of heart sounds sensor and restructure a wireless voice detection device by using our own design of a " detection device with double auscultative head and two kinds of heart sounds". The detection device with double auscultative heads and the diagram of wireless heart sound detection device is shown in figure 3. It can extract heart sound signal effectively by two paths. The wireless heart sound detection device is constituted by Signal regulating circuit, Data acquisition system ADuC812, external RAM and Bluetooth module behind of this device. Because the ADC can only achieve one polarity analog-to-digital conversion in the ADuC812, moreover, the biggest input signal range is 0 ~V DD. So it is essential to use the modulation circuit to process external simulation signal and convert it to the signal as the ADC requires. Besides, the circuit for regulating signal can also be used to amplify the heart sound signal and converse the external input of heart signal to the level of ADC input level range. After being processed by ADuC812, the data is stored in external RAM and sent out by been encapsulated into packets. The receiver puts the heart sound signal data into the computer after connecting the Bluetooth modules in the USB interface of computer. The specific process is as follows: (1) Put the wireless heart sound detection device on the heart. When you success to choose the best auscultation place, press the pressure switch button hard. Then the detection device will convert the heart throb process to electrical impulses heart sound signal. (2) After using the circuit that regulates signal to amplify, filter and isolate the heart sound signals, which will be converted into data acquisition system called chip ADuC812 and go through multiple switch, A/D conversion (sampling, keep), data storage, then been encapsulated into packets and sent to Bluetooth module. (3) First, initialize Bluetooth devices with USB interface by computer, then you can search and connect Bluetooth module. After being sent to the computer through Bluetooth module, the data will be processed by heart sounds analyzer based on LabVIEW. The working power that the Wireless device for wireless heart sound detection device requires is 3.3V, -5V, +5V. In order to reduce the volume of the system and carry it easily, we often use two stars of button batteries. Therefore, on the one hand, we use power module that can rise voltage and choose MAX756 power module meeting some needs, whose input voltage range is 2V ~V out, and output voltage is 3.3V or 5V. Besides, if use reversed device such as ICL7660, it can convert positive voltage 5V to negative 5V. The test device has the characteristics of 10 ~ 1000 times magnification, automatic adjustment and overload capacity of 50 times. Its frequency response is 0.1 ~1300 Hz. The frequency of heart sounds signal range is 10HZ ~ 400HZ, but according to the sampling theorem, the actual sampling frequency can only be 800Hz and above. Figure 2 the periodicity of the heart sound signals and the frequency characteristics of s1, s2 So, we designed a strategy for this virtual instrument s 2. HEART SOUND ANALYZER 2.1. Heart sounds acquisition subsystem

3 The main function of heart sounds acquisition subsystem is that it records heart sounds based on the acquisition format specified by users and stores data in the specified path and it also shows acquisition waveform at the same time. Comparative experiments have found that in the same noisy environment to use this wireless heart sounds detection device, the lower frequency is, the larger interference generated by noise is. And with the improvement in the acquisition frequency, the first heart sounds and the second recorded become more and more obvious. Considering that the higher frequency will consume more memory and the signal of heart sounds is clear enough at KHz, we use KHz as default (Cheng at el., 2010). According to the analysis in last section, we respectively set sampling rate, the channel number and sampling bits value of 11025, 2, and 16. Also users can change these values based on the actual sampling conditions. We configure paths to store data in saving path of the file, and then click the button named heart sound collection to start collecting. Finally we end acquisition by clicking the button named stop the collection, and the heart sounds will be saved automatically in the path we have set. Acquisition process can be real-time observed in the display control. Due to the continuous acquisition process, we can t transmit all the data at one time. So establishing a while circulation is necessary. The loop won t end until we click the button named stop the collection, Sound Input Stop sub VI will stop sound input device collecting and Sound Input Clear sub VI will clear buffers. Finally we use Sound File Close sub VI to close the open file. Fig. 4. The program diagram of heart sounds acquisition subsystem 2.2. Heart sounds de-noising subsystem Heart sound de-noising subsystem based on wavelet analysis is constituted by Type Selection of Wavelet Generating Function, Determination of Decomposition Layers and Threshold Setting. In order to make the denoising process more flexible, the system set each parameter as adjustable and put the scheme using db3 wavelet to de-noise in five layers as default Settings, soft threshold denoising is default and unchangeable. "Zeros percentage indicates the ratio the coefficient each layer is set to 0 in total coefficient. Systems will calculate the denoising threshold of each layer in accordance with the parameters. After setting the various parameters, users can press the "Wavelet de-noising" button to start denoising. The de-noised heart sounds signal will be displayed on the display window (Bulgrin at el., 2003; Debbal at el., 2004) Heart sounds time-domain analyzing subsystem Figure 3 The Wireless heart sound detection device The program diagram of heart sounds acquisition subsystem is shown in figure 4. We mainly use LABVIEW Sound File Write Open sub VI, Sound Format parameter setting cluster, Sound Input Configure sub VI, Sound Input Read sub VI, Sound Input Stop sub VI, Sound Input Clear sub VI and Sound File Close sub VI. Firstly, we use Sound File Write Open sub VI to open file specified by saving path of the file and Sound Input Configure sub VI will set the sound format according to Sound Format parameter setting cluster. Then Sound Input Read sub VI is about to read data from sound input device (here is soundcard) and transfer them to the oscilloscope to real-time display and to the Sound File Write sub VI to write them into files. Heart sounds time-domain analyzing subsystem is mainly to calculate heartbeat frequency and the interval between heart sounds. The concrete method to determine the beginning and ending of the first heart sound and the second heart sound is: (1) Suppose not aliasing, we conduct the second sampling to heart sounds signal in order to reduce the amount of work in data calculation and process rapidly in the follow up. As we see in figure 7(A), (a) is the original heart sound signal sx and (b) is the heart sound signal sy after the second sampling, the amount of data is significantly reduced. (2) Calculate the energy spectrum of the signal after the second sampling. That is: p(i) = sy(i) 2 i = 0,1,2 as (c) shows in figure 7(A). (3) As for heart sound energy spectrum that is obtained, we use the layering method of empirical model to extract its envelope. Some peaks of the envelope exists the condition that they become small in short time. in order to determine the beginning and ending of the first heart sound and second

4 heart sound, according to the features of s1 and s2, set the average value of the envelope as threshold and do not consider the change during less than 20ms for continuous time intervals, then we can get the normalized energy envelope, as (d) shows in figure 7(A). (4) Feed the above information back to original heart sound signal, we can definitely obtain the beginning and ending of the first and second heart sound, as (e) shows in figure 7(A), so we can accurately segment heart sound signal of a circle. Calculate the average heart sound period by extracting heart sound signal of 3 circles, and then calculate the heartbeat number. So, it can become a heart rate meter too. (5) According to the features of heart sound, for the individual, short duration, no transitional, and energy to become suddenly big and no repetitive signals, regard them as sudden interference caused by the stethoscope s movement, so we set the threshold of energy to remove the blowout high-energy section. Because the calculation of this subsystem is complex, the program frame will be very large if we use the control oflabview to fulfill it. For the sake of simplicity, we still use Matlab script node to simplify the design. Through Matlab which has a powerful ability in numerical calculation, we will be able to complete the extraction of signal envelope, normalization, the calculation of heartbeat and heart sounds interval, and then transport the calculation results to the front panel display window. The final program diagram will be very concise, as is shown in figure Heart frequency domain analysis subsystem In the diagnosis of heart sound, abnormal heart sound often easily to be observed, because they contain complex and obvious frequency domain components in frequency domain. So it is meaningful to analysis the frequency domain of heart sounds. The subsystem to accomplish heart sound of frequency domain analysis mainly includes FFT spectrum analysis, wavelet decomposition and Wigner distribution analysis. When users use the subsystem to analysis the frequency domain of heart sound, they no longer need to do the boring theory research, learn programming language, and look up material and code. They just need to click one or two buttons and then can achieve complex numerical analysis, which is right-hand man in the heart sound of frequency domain analysis. The users just click the button of "before the de-noising FFT spectrum" or "after the de-noising FFT spectrum" can get the before and after de-noising spectrums respectively, which are convenient to comparison the effect of de-noising in frequency domain. The photo shows a sampling signal spectrum before de-noising. From the photo we can see that the frequency range of heart sound signal mainly distributed in 0 ~ 150Hz. The LabVIEW provides FFT spectrum function node which convenient software developer calls (be shown as in the figure 6). This function calculates the average FFT spectrum of time signal, returns FFT values as "range" and "phase". Figure 6 FFT spectrum function Figure 5 the diagram of time-domain analyzing subsystem program The subsystem calculates sampling signal s heart beat frequency and heart sounds interval correctly. Its runningtime is 6s. We use 50 cases of heart sound signals which are sampled after denoising to test the performance of the subsystem; the accuracy can reach to 99%. Otherwise, we use another 50 cases of sampled signals without being de-noised to test this system; the accuracy can reach to 85%. As these tests show, the subsystem represents good stability and anti-noise ability. For the heart sound signals, we always want to see the heart sound spectrum distribution in some specific points or periods of time. Wigner - Ville algorithm can solve the problem for better. It is one of characteristics in the heart sound analyzer (L.Wang at el., 2010). 1 * 1 1 -jw WS t = S - + k k t t Sk t t e dt 2 (2) 2 2 After users choose the "wig2" or "wig2c" in distribution type selection drop-down list,then press ok,the system will output Wigner-Ville distribution figure and meanwhile the heart sounds waveform will be shown in the next cycle time domain, which is convenient for the users to observe in contrast. There are graphical tools to choose beside every figure, users can use these tools to make it larger or smaller, move it and observe one point s accurate value and so on. In this sub module, we combine Matlab with LabVIEW to simplify design. Matlab provides Wig2 and Wig2C to compute Wigner -Ville distribution, and the Wig2C uses Choi - William filter to cross the interference item Wigner spectrum. As it is time-consuming to compute Wigner-Ville s distribution, we are not going to compute all Wigner- Ville s sampling period distribution and only take one of the cycles. Which cycle is to take? The starting of this cycle comes from the course of TAB of Wavelet de-noising. The intersection of two yellow lines in the display window is shown. The values

5 of the course can be changed by dragging mouse. It is one of the Waveform figure s attribute, which can be obtained by waveform figure s attribute node. The system will see the cursor as the starting point, and then take about a cycle after it to calculate the distribution of Wigner Ville. A cycle of heart sound signal Wigner - Ville distribution results are shown in figure 7(B). sound S2. t 23 is the interval of the second heart sound S2 and the third heart sound S3. t 34 is the interval of the third heart sound S3 and the fourth heart sound S4. t 41 is the interval of the fourth heart sound s4 and the first heart sound S1. c 1i, c 2i, c 3i, c 4i, c ni is means the amplitude parameters of the first heart sound s1, the second heart sound S2, the third heart sound S3, the fourth heart sound S4 and the noise S N.T is the total number of sampling data about the periodic heart sound. When use it, you must adjust relevant paremeters according to the needs of users and change the standard templates into the user s templates thus structured the heart sound meeting the user s demand. Whether it is normal or abnormal heart sound, it can be used as the standard heart sound signal. (A) Fig. 7. The determination of the beginning and ending of the first and second heart sound and a cycle of heart sound signal s Wigner - Ville distribution results figure 2.5. Subsystem of heart sound signal generator The realization of the heart sound signal generator is that: At first, A standard cycle of heart sound signal must be decomposed into six parts ---- the first heart sound S1,the second heart sound S2,the third heart sound S3,the fourth heart sound S4 and the interval of heart sound t xy, the noises N. Second, the six parts of the relevant parameters should be extracted and constitute a set of standard templates, the templates can be expressed as follows: T s (f ) (c s (f ) t c s (f ) t T 1i i i 1 c s (f ) t c s (f ) t c s 3i i Ni N In which, f1, f2, f3, f4 is means the frequency parameter of the first heart sound S1, the second heart sound S2, the third heart sound S3 and the fourth heart sound S4. t 12 is the interval of the first heart sound S1 and the second heart ) (3) (B) Fig. 8. Heart sound signal generator s block diagram of the process and Front panel Generally, S3 and S4 can not be heard, because of sampled heart sound signal S3and S4 are very weak. So the system only involves heart sounds signal S1, S2 and noise. In fact, the theory of generating S3, S4 is the theory of generating as S1 and S2. Based on this process, heart sound signal generator s block diagram is shown in Figure 6(A) and its front panel is shown in Figure 8(B). Figure 9 is the heart sounds generated by a set of arbitrary parameters.

6 actively involved in this research area and contribute to its growth and advancements. ACKNOWLEDGEMENT: This work is supported by the National Natural Science Foundation of China (Grant No , , ) and the National Natural Science Foundation of Jiangsu Governorate (Grant No.BK ). REFERENCES AND NOTES Figure 9 normal heart sounds generated with the arbitrary 3. CONCLUSION parameters The instrument uses the wireless pressure sensors and PC control as hardware, based on the LABVIEW development platform and combines with MATLAB thus develops a set of heart sound analysis virtual instrument which includes many functions such as acquisition, noise reduction, analysis, and automatically generation of heart sound. In order to make the instruments more flexible when used, each parameter is can be adjustable. This research on heart sound analysis instrument has grown from the fact that systems like described above is a reality that is becoming part of our lives and can not be avoided. Now with advanced time and frequency domain signal processing methods, the development of a modern device, which is independent of the physician s hearing acuity with enhanced signal acquisition, processing, display, and replay capabilities, has become a feasible reality. Witnessing this technical proliferation opens up new exciting opportunities for biomedical engineers to be Bulgrin, J. R., Rubal, B. J., Thompson, C. R., and Moody, J. M.,Comparison of short-time Fourier, Wavelet and time domain analyses of intercardiac sounds. Biomedical Sciences Instrumentation. 29, 465 (2003). Cheng Xiefeng, Ma Yong, Liu Chen, et al. Research on heart sound identification technology [J].Sci China Inf Sci, 2012, ( 55 ) : Cheng X F, Ma Y, Zhang S B. Three-step identity recognition technology using heart sound based on information fusion, Chinese Journal of Scientific Instrument(in Chinese). 31, 1712 (2010). Debbal, S. M., and Bereksi-Reguig, F., Analysis of the second heart sound using continuous wavelet transforms. Journal of Medical Engineering & Technology. 28, 151(2004). Jiang ZW, Choi S. A Cardiac Sound Characteristic Waveform Method or In-home Heart Disorder Monitoring with Electric Stethoscope [J]. Expert Systems with Applications. 31,286 ( 2006). L. Wang, K. Kalyanasundaram, M. Stanacevic, and P. Gouma, Nanosensor Device for Breath Acetone Detection. Sensor Lett. 8, 709 (2010). National Instruments Corporation. LabVIEW User Guide. National Instruments. (2008). Taner Topal. Hüseyin Polat. Software Development for the Analysis of Heartbeat Sounds with LabVIEW in Diagnosis of Cardiovascular Disease. Springer Science & Business Media. 32,409 (2008).