A cardiac sound characteristic waveform method for in-home heart disorder monitoring with electric stethoscope Zhongwei Jiang * , Samjin Choi Department of Mechanical Engineering, Yamaguchi University, 2-16-1, Tokiwadai, Ube, Yamaguchi, 755-8611, Japan Abstract An analytical model based on a single-DOF is proposed for extracting the characteristic waveforms (CSCW) from the cardiac sounds recorded by an electric stethoscope. Also, the diagnostic parameters [T1, T2, T11, T12], the time intervals between the crossed points of the CSCW and an adaptive threshold line (THV), were verified useful for identification of heart disorders. The easy-understanding graphical representation of the parameters was considered, in advance, even for an inexperienced user able to monitor his or her pathology progress. Since the diagnostic parameters were influenced much by a THV, the FCM clustering algorithm was introduced for determination of an adaptive THV in order to extract reliable diagnostic parameters. Further, the minimized J m and [v 1 , v 2 , v 3 , v 4 ] could be also efficient indicators for identifying the heart disorders. Finally, a case study on the abnormal/normal cardiac sounds is demonstrated to validate the usefulness and efficiency of the cardiac sound characteristic waveform method with FCM clustering algorithm. NM1 and NM2 as the normal case have very small value in J m (!0.02) and the centers [v 1 , v 2 , v 3 , v 4 ] are about [0.1, 0.1, 0.8, 0.4]. For abnormal cases, in case of AR, its J m is very small and the values of [v 1 , v 3 , v 4 ] are very high comparing to the normal cases. However, in cases of AF and MS have very big values in J m (O0.38). q 2005 Elsevier Ltd. All rights reserved. Keywords: Electric stethoscope auscultation; Cardiac sound characteristic waveform (CSCW); Primary health care; Automatic data processing; Heart murmurs disorder 1. Introduction The death due to heart disease in the world became to the second mortality after the stroke (cerebrovascular accident) since 1985. Furthermore, based on a medical certificate of death the majority of deaths caused by cardiac diseases are of the heart failure and coronary heart disease. However, except the identified diseases due to the cardiac diseases still 30% deaths are of unknown causes. Some of them might be due to the cardiac diseases. If life-style related diseases could not be monitored continuously during a long period, some cardiac diseases like coronary heart disease, angina pectoris and myocardial infarction might be difficult to be diagnosed appropriately and detected in an early step. In the recent year, the high concern about health manage- ment and medical welfare makes the rapid development of home medical instruments for health care and diagnosis in daily life. Stethoscopes, in addition to other health care instruments such as weight scale, a clinically thermometer and a sphygmomanometer, have come into wide use for inexperi- enced users. Since the stethoscope could auscultate the respiratory sounds, lung sounds as well as cardiac sounds, and screen the most cardiorespiratory disorders and diseases, it might become a cheap and efficient home health care instrument in the near future. Recently the stethoscope has been used for auscultating embryonic (including fetal) cardiac sounds from pregnant mothers or for health management of pets in home. However, using the stethoscope to screen human’s disorder needs a long-term practice and experience. Even for a well-trained young cardiologist to auscultate and diagnose cardiac diseases several years’ clinic experience is required. Actually a well-experienced cardiologist could hear out the pathologic heart murmur very sensitively but it is so difficulty to an inexperienced or non-clinical experience person. Therefore, if the heart sound could be recognized or diagnosed with the support of computer software technique, the above problems will be solved and the stethoscope may be taken advantage of as a high-quality home medical and health care instrument. The researches on diagnosis of heart diseases were concentrated around in the 1970s and there are a lot of results reported (Yoshimura, 1973; Machii, 1972; Yokoi, 1974; Iwata, Expert Systems with Applications 31 (2006) 286–298 www.elsevier.com/locate/eswa 0957-4174/$ - see front matter q 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2005.09.025 * Corresponding author. Tel.: C81 836 85 9137; fax: C81 836 85 9137. E-mail addresses: [email protected] (Z.W. Jiang), b3678@yamaguchi- u.ac.jp (S. Choi)
13
Embed
A cardiac sound characteristic waveform method for in … · A cardiac sound characteristic waveform method for in-home heart disorder monitoring with electric stethoscope ... 1977;
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
A cardiac sound characteristic waveform method for in-home heart
disorder monitoring with electric stethoscope
Zhongwei Jiang *, Samjin Choi
Department of Mechanical Engineering, Yamaguchi University, 2-16-1, Tokiwadai, Ube, Yamaguchi, 755-8611, Japan
Abstract
An analytical model based on a single-DOF is proposed for extracting the characteristic waveforms (CSCW) from the cardiac sounds recorded
by an electric stethoscope. Also, the diagnostic parameters [T1, T2, T11, T12], the time intervals between the crossed points of the CSCW and an
adaptive threshold line (THV), were verified useful for identification of heart disorders. The easy-understanding graphical representation of the
parameters was considered, in advance, even for an inexperienced user able to monitor his or her pathology progress. Since the diagnostic
parameters were influenced much by a THV, the FCM clustering algorithm was introduced for determination of an adaptive THV in order to
extract reliable diagnostic parameters. Further, the minimized Jm and [v1, v2, v3, v4] could be also efficient indicators for identifying the heart
disorders. Finally, a case study on the abnormal/normal cardiac sounds is demonstrated to validate the usefulness and efficiency of the cardiac
sound characteristic waveform method with FCM clustering algorithm. NM1 and NM2 as the normal case have very small value in Jm (!0.02)
and the centers [v1, v2, v3, v4] are about [0.1, 0.1, 0.8, 0.4]. For abnormal cases, in case of AR, its Jm is very small and the values of [v1, v3, v4] are
very high comparing to the normal cases. However, in cases of AF and MS have very big values in Jm (O0.38).
q 2005 Elsevier Ltd. All rights reserved.
Keywords: Electric stethoscope auscultation; Cardiac sound characteristic waveform (CSCW); Primary health care; Automatic data processing; Heart murmurs
disorder
1. Introduction
The death due to heart disease in the world became to the
second mortality after the stroke (cerebrovascular accident)
since 1985. Furthermore, based on a medical certificate of
death the majority of deaths caused by cardiac diseases are of
the heart failure and coronary heart disease. However, except
the identified diseases due to the cardiac diseases still 30%
deaths are of unknown causes. Some of them might be due to
the cardiac diseases. If life-style related diseases could not be
monitored continuously during a long period, some cardiac
diseases like coronary heart disease, angina pectoris and
myocardial infarction might be difficult to be diagnosed
appropriately and detected in an early step.
In the recent year, the high concern about health manage-
ment and medical welfare makes the rapid development of
home medical instruments for health care and diagnosis in
daily life. Stethoscopes, in addition to other health care
0957-4174/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.
Hathaway, Sabin, Tucker, Bezdek & Pal, 1992) is introduced
for determining the threshold values.
Considering a data series of ZZ{z1,z2,.,zj,.,zn} to be
clustered into C groups, where zjZ[z1,., zk,., zc]j, the center
vi of the ith clustered group can be defined by
vi Z
Pn
jZ1
ðwi;jÞmzk;j
Pn
jZ1
ðwi;jÞm
; i Z 1; 2;.;C; (3)
where wi,j is the degree of belongingness specified by a fuzzy
membership grades between 0 and 1 with the constrainsPCiZ1
wi;jZ1, cjZ1,.n; vi is the cluster center of data group i;
m2[1,N] is a weighting exponent. The Euclidean distance di,j
between the i-th cluster center and the jth data point is then
defined as
di;j Z jjviKzk;jjj; (4)
The cost function for FCM clustering is given by
JmðW ;VÞZXC
iZ1
XN
jZ1
ðwi;jÞmðdi;jÞ
2; (5)
where WZ{wi,j} and VZ{vi}. The algorithm to calculate the
cluster centers {vi} and the membership matrix {wi,j} works
iteratively by minimizing Eq. (5) with an obtained wi,j, which is
calculated by the Euclidean distance obtained in the previous
step using the following equation
wi;j Z1
PCkZ1
ðdi;j=dk;jÞ2=ðmK1Þ
: (6)
The performance of FCM clustering method depends on the
initial membership grade values. It is recommended to run the
algorithm for several times, each starting with different values of
membership grades. In following analysis, the weighting exponent
m is set at 2 and the number of iteration g and the cost function Jm
are used for evaluation of the performance of the FCM algorithm.
Thinking to cluster the parameters [T1, T2, T11, T12] into
four groups, each cluster center vi can be obtained easily by
following the FCM clustering algorithm if letting [z1, z2, z3, z4]jas [T1, T2, T11, T12]j. Imitating the scattergram, the data [z1,
z2] and [z3, z4] are plotted in a two-dimensional way as shown
in Fig. 11. Thereby, the cluster centers can be expressed by
two-dimension vectors [v1, v2] and [v3, v4], which are marked
as A and B in Fig. 11.
Table 1
The obtained cluster centers [v1, v2, v3, v4] and minimized cost function Jm with respect to each THV from a normal cardiac sound signal
THV (%) Jm v1 (s) v2 (s) v3 (s) v4 (s)
10 0.62866 0.1300 0.1170 0.534 0.384
20 0.14034 0.1140 0.1020 0.749 0.392
30 0.00932 0.0988 0.0768 0.751 0.348
40 0.00938 0.0844 0.0654 0.751 0.336
50 0.00919 0.0700 0.0529 0.752 0.326
60 0.00911 0.0556 0.0399 0.752 0.315
70 1.16464 0.0494 0.0387 0.842 0.374
Z. Jiang, S. Choi / Expert Systems with Applications 31 (2006) 286–298 293
4.3. Determination of the threshold values by FCM clustering
method
Table 1 shows the obtained cluster centers from a normal
cardiac sound signal when the THV is varied from 10 to 70%.
The cost function Jm shows bigger values at THVZ[10, 20,
70%] than those at THVZ[30, 40, 50, 60%]. This means the
data are scattering wider at a big value of the cost function Jm
than at a small value. Confirming the fact, all of the parameters
[T1, T2] and [T11, T12] obtained at each THV from 10 to 70%
with 10% step are plotted together in Fig. 12(a) and (b). In
contrast, the data obtained only at small value of Jm, i.e.
THVZ[30, 40, 50, 60%], are plotted in Fig. 12(c) and (d).
From Table 1 and Fig. 12, it could firmly conclude that the
THV selected at the minimized cost function produces a
reliable series of the parameters [T1, T2, T11, T12]j.
The parameters [T1, T2, T11, T12]j are the crossed points of
the characteristic curve on the threshold line, and they are
stored into a series in turn from the first crossed point.
Fig. 12. Scattergrams of [T1, T2] and [T11, T12]; (a) and (b) are the plots at all THV
40, 50, 60%].
This means there are four possibilities in combination as
follows,
T1j T2j T11j T12j
T12j T1j T2j T11j
T11j T12j T1j T2j
T2j T11j T12j T1j
266664
377775
Thereby, the cluster centers {vi} might have also four
combinations corresponding to each data set {T}. However,
the data set {T} could be reduced to cases, i.e. [T1, T2, T11,
T12] and [T2, T11, T12, T1] or [T11, T12, T1, T2] and [T12,
T1, T2, T11], if two-dimensional plots [z1, z2] and [z3, z4] are
considered. Suppose, if the first crossed point is starting at T2,
the sequential order of the data will be [T2, T11, T12, T1]j.
Rearranging the data of Fig. 12 into a series of [T2, T11, T12,
T1], the scattergrams respect to [T2, T11] and [T12, T1] are
plotted in Fig.13(a) and (b). Further, the results obtained only at
s in Table 1, (c) and (d) are the results obtained only at the suitable THVZ[30,
Fig. 13. Scattergrams obtained by rearranging the data series of Fig. 12; (a) and (b) are the plots at all THVs in Table 1, (c) and (d) are the results obtained only at the
suitable THVZ[30, 40, 50, 60%].
Z. Jiang, S. Choi / Expert Systems with Applications 31 (2006) 286–298294
the minimized cost function, i.e. at THVZ[30, 40, 50, 60%]
are plotted in Fig. 13(c) and (d).
In general case, the centers [v3, v4], corresponding to [T11,
T12], is larger than the centers [v1, v2], corresponding to [T1,
T2]. Therefore, the data sequence of [T1, T2, T11, T12] or [T2,
T11, T12, T1] can be identified by comparing the center values.For example, if v3 Ov2 the data sequence will be [T1, T2, T11,
T12] and if v2Ov4, the data sequence will [T2, T11, T12, T1].
4.4. Results and discussions
4.4.1. Case of normal cardiac sounds
Two normal examples, which data were shown in Fig. 6,
were used first to explain the efficiency of the FCM clustering
method. Fig. 14 shows the minimized value of the cost
Fig. 14. Plots of the minimized cost function Jm and cluster centers [v1, v2, v3
function Jm and the cluster centers [v1, v2, v3, v4] as a function
of the threshold value (THV). The dotted lines indicate the
adaptive area of THV for calculating the parameters [T1, T2,
T11, T12]. It is shown that the lowest Jm are along the regions
THVsZ20–40% for case NM1 and 15–60% for case NM2.
The mean value of Jm along THVsZ20–40% is about 0.02212
at case NM1 and is 0.00657 along THVsZ15–60% at case
NM2. Referring to the plots of the cluster centers [v1, v2, v3,
v4], it is found that they almost keep constant in these regions.
The mean values of the centers [v1, v2, v3, v4] are [0.06954,
0.07527, 0.75700, 0.37757] s, respectively, for case NM1, and
[0.08969, 0.06917, 0.76980, 0.33950] s for case NM2. And
the corresponding cardiac cycle can be calculated by the value
v3, i.e. CyZ60/v3Z79.2602 beats/min for case NM1 and
94.6372 beats/min for case NM2.
, v4] as a function of the threshold value (THV) for normal heart sounds.
Fig. 15. The scattergrams of parameters [T1, T2] and [T11, T12], which are determined within the region of THVsZ20–40% and 15–60% for NM1 and NM2,
respectively.
Z. Jiang, S. Choi / Expert Systems with Applications 31 (2006) 286–298 295
Next, one uses the threshold values (THVs) of 20–40% by
each 10% step for case NM1 and THVs of 15–60% by each
10% step for case NM2 to calculate the parameters [T1, T2]
and [T11, T12], and plots the obtained results of case NM1 (6)
and NM2 ( ) in Fig. 15. It shows that these data are all grouped
quite well at the adaptive THVs. It could be concluded that the
FCM clustering method is an available tool for obtaining the
Fig. 16. Plots of the minimized cost function Jm and cluster centers [v1, v2, v3, v
reliable parameters [T1, T2] and [T11, T12] from the heart
sound characteristic waveforms.
4.4.2. Case of abnormal cardiac sounds
Some more examples are shown for verification of the FCM
clusteringmethodwhen it is applied to abnormal heartmurmurs.
In contrast, the abnormal sound data for testing are of the atrial
4] as a function of the threshold value (THV) for cases of AF, MS, and AR.
Fig. 17. Scattergrams of abnormal heart sounds AF, MS and AR, which parameters [T1, T2] and [T11, T12] are calculated at the region of THVsZ[16–46%, 45–
66%, and 10–22%], respectively.
Fig. 18. Summary of Figs. 14 and 17, which could be used as the indicators to identify the heart disorders.
Z. Jiang, S. Choi / Expert Systems with Applications 31 (2006) 286–298296
Fig. 19. Data processing procedure of the proposed CSCW method with the
data clustering technique.
Z. Jiang, S. Choi / Expert Systems with Applications 31 (2006) 286–298 297
fibrillation/flutter (AF) as shown in Fig. 7, the mitral stenosis
(MS) in Fig. 8 and the aortic regurgitation (AR) in Fig. 9.
To determine the threshold values (THVs), the FCM
clustering algorithm is introduced to these cases. The cluster
centers [v1, v2, v3, v4] and the corresponding minimized cost
function Jm as a function of THV are described in Fig. 16.
The adaptive THVs should be selected at the lowest values
of Jm, so one can determine the values THVsZ16–46% for
case of AF, 45–65% for case MS, and 10–22% for case AR
easily from Fig. 16. Therefore, the parameters [T1, T2] and
[T11, T12] for each abnormal cases are obtained along the
adaptive THVs and plotted onto the scattergrams as shown in
Fig. 17.
Fig. 17(a) and (b) are of case AF, (c) and (d) for MS, and (e)
and (f) for AR, respectively. In case of AF, the parameters
looked almost in normal except the parameter T11, which is
distributed along a wide region, say about 0.4–1.0 s (Fig. 17b).
This indicates that the arrhythmia can be identified by the
parameter T11. Next, looking the case of MS, it is found that
the parameters [T1, T2] and [T11, T12] are scattered widely
over whole the region as shown in Fig. 17(c) and (d). This
means that MS sound could be easily recognized by the
scattergrams of the parameters. As for the case of AR, the
parameters T1 and T12 (Figs. 9 and 17e and f) have longer time
duration than the ones in normal cardiac sound case although
their scattergrams are grouped well and concentrated in a very
small areas. In case of the diagnosis of AR, the time durations
of T1 and T12 could be good candidates.
4.4.3. Detection of abnormality in heart sounds
Fig. 18 shows the summary of above obtained results. The
values Jm and [v1, v2, v3, v4] in the figure are the averages along
the adaptive THV regions as shown by two dotted lines in Figs.
14 and 16. Also their standard derivations are expressed by the
error bars. It shows obviously that the values of Jm and [v1, v2,
v3, v4] could be efficient indicators for distinguishing the
normal and abnormal heart sounds. From the figure, as an
example, the normal heart sounds (NM1 and NM2) have very
small value in Jm, lower than 0.02 and the centers [v1, v2, v3,
v4], or the average time durations of [T1, T2, T11, T12], are
about [0.1, 0.1, 0.8, 0.4] s. However, for an abnormal heart
sound, at least one of these values are extremely higher than the
one in normal case. For example, abnormal sounds of AF and
Ms have almost the same values of [v1, v2, v3, v4] as the normal
case, but their cost functions Jm are extremely high almost over
0.4 which is about twenty times higher comparing with the
normal case NM1. As for abnormal case of AR, its Jm is very
small but the values of [v1, v3, v4] are very high and they are
beyond the normal ranges. Furthermore, comparing these data
which have higher values of Jm, such as the examples of the
plots in Figs. 7–9 and 17, it is found that the value Jm of the
minimized cost function is a kind of scattering scale of the data,
which could be used as a powerful indicator for identification
of heart murmurs.
Finally, the data processing procedure based on the
proposed cardiac sound characteristic waveform method with
the data clustering technique is described in Fig. 19. Appling
the proposed method into the practical use still needs an
amount of clinic tests and construction of each personal
database. It will be continued in our future works.
5. Conclusions
In this paper, a novel cardiac sound analysis method for in-
home cardiac disorder detection and monitoring with a simple
electric stethoscope was described. An analytical model based
on a single-DOF was proposed for extracting their character-
istic waveforms from the cardiac sounds so that the heart
sounds could be easily treated by computer for screening heart
disorder.
The diagnostic parameters were defined by the time
durations on and between the first and second sounds, which
was verified useful for identification of heart disorders. The
easy-understanding graphical representation of the parameters
was considered, in advance, even for an inexperienced user
able to monitor his or her pathology progress. Since the
diagnostic parameters were influenced much by a threshold
valve (THV), the Fuzzy C-means (FCM) clustering algorithm
was, thereon, introduced for determination of an adaptive THV
in order to extract reliable diagnostic parameters. Further, the
minimized cost function and the cluster centers could be also
efficient indicators for identifying the heart disorders.
Z. Jiang, S. Choi / Expert Systems with Applications 31 (2006) 286–298298
Finally, a case study on the abnormal/normal cardiac sounds
is demonstrated to validate the usefulness and efficiency of the
cardiac sound characteristic waveform method with FCM
clustering algorithm.
References
Adolph, R. J., Stephens, J. F., & Tanaka, K. (1970). The clinical value of
frequency analysis of the first heart sound in myocardial infarction.
Circulation XLI, 41, 1003–1014.
Akay, M. (1994). Automated noninvasive detection of coronary artery disease
using wavelet-based neural networks. Intelligent Engineering System
Artificial Neural Network, 14, 517–522.
Ammash, N. M., & Warnes, C. A. (2001). Ventricular septal defects in adults.
American College of Physicians–American Society of Internal Medicine,
135, 812–824.
Barschdorff, D. (1995). Phonocardiogram signal analysis in infants based on