Content uploaded by Miguel A Becerra
Author content
All content in this area was uploaded by Miguel A Becerra on Nov 30, 2018
Content may be subject to copyright.
Exploratory Study of the Effects
of Cardiac Murmurs
on Electrocardiographic-Signal-Based
Biometric Systems
M. A. Becerra1,2(B
), C. Duque-Mej´ıa1, C. Zapata-Hern´andez1,
D. H. Peluffo-Ord´o˜nez3, L. Serna-Guar´ın4, Edilson Delgado-Trejos4,
E. J. Revelo-Fuelag´an5, and X. P. Blanco Valencia3
1Instituci´on Universitaria Pascual Bravo, Medell´ın, Colombia
migb2b@gmail.com
2Universidad de Medell´ın, Medell´ın, Colombia
3SDAS Research Group, Yachay Tech, Urcuqu´ı, Ecuador
4Instituto Tecnol´ogico Metropolitano, Medell´ın, Colombia
5Universidad de Nari˜no, Pasto, Colombia
http://www.sdas-group.com
Abstract. The process of distinguishing among human beings through
the inspection of acquired data from physical or behavioral traits is
known as biometric identification. Mostly, fingerprint- and iris-based bio-
metric techniques are used. Nowadays, since such techniques are highly
susceptible to be counterfeited, new biometric alternatives are explored
mainly based on physiological signals and behavioral traits -which are
useful not only for biometric identification purposes, but may also play
a role as a vital signal indicator. In this connection, the electrocardio-
graphic (ECG) signals have shown to be a suitable approach. Nonethe-
less, their informative components (morphology, rhythm, polarization,
and among others) can be affected by the presence of a cardiac pathol-
ogy. Even more, some other cardiac diseases cannot directly be detected
by the ECG signal inspection but still have an effect on their waveform,
that is the case of cardiac murmurs. Therefore, for biometric purposes,
such signals should be analyzed submitted to the effects of pathologies.
This paper presents a exploratory study aimed at assessing the influence
of the presence of a pathology when analyzing ECG signals for imple-
menting a biometric system. For experiments, a data base holding 20
healthy subjects and 20 pathological subjects (diagnosed with different
types of cardiac murmurs) are considered. The proposed signal analysis
consists of preprocessing, characterization (using wavelet features), fea-
ture selection and classification (five classifiers as well as a mixture of
them are tested). As a result, through the performed comparison of the
classification rates when testing pathological and normal ECG signals,
the cardiac murmurs’ undesired effect on the identification mechanism
performance is clearly unveiled.
c
Springer Nature Switzerland AG 2018
H. Yin et al. (Eds.): IDEAL 2018, LNCS 11314, pp. 410–418, 2018.
https://doi.org/10.1007/978-3-030-03493-1_43
Exploratory Study of the Effects of Cardiac Murmurs 411
Keywords: Biometric identification ·Cardiac murmur
Electrocardiographic signal ·Signal processing
1 Introduction
The biometric identification is the verification of the identity of the person based
on characteristics of his body, either by features or signs. Biometric identification
has been a subject of great interest, and in recent years, these systems have
been seen in many of the daily tasks using fingerprint or iris. The applications of
biometric identification are multiple, ranging from the identification of diseases
for diagnostic purposes to the identification of individuals for security purposes,
which has improved in many aspects, the privacy of people and the improvement
in the diagnosis of the diseases avoiding mistakes due to identity confusion.
One of the most used identification techniques is the fingerprint, which is
still a very efficient method since each has different morphological characteris-
tics and is a trait of easy acquisition [15]. Nevertheless, it is very vulnerable to
counterfeiting [12,14]; therefore it is applied supervised way, but at the same
time, different techniques have been developed to avoid these identification sys-
tems, which are falsifies using different materials that allow acquiring and use
this fingerprint illegally [20].
To improve the shortcomings of conventional biometrics [22] such as finger-
print [11] and iris [5], the use of the electrocardiogram (ECG) as a method of
biometric identification has been proposed [15,16,19]. ECG signals are reliable
in comparison with conventional biometrics because they generate greater secu-
rity and manage to meet many of the requirements necessary to become an
ideal biometric system [17]. ECG has important characteristics such as the size
of the heart and spatial location, different subjacent electrical and mechanical
dynamics that become unique, among others. This signal has uniqueness, perma-
nence, universality, and evasion. In comparison with other biometrics, the ECG
has the inherent capacity of life detection for identification [4]. Moreover, the
ECG signals can be acquired by three different acquisition methods defined by
[3] as follows: (i) in the person (use invasive equipment is used, designed to be
used within the human body), (ii) on the person (signals acquired by electrodes
attached to the skin), and (iii) off the person (minimal contact on the skin or
without contact), an example is to acquire the signal through fingers [8].
Currently, there have been multiple studies of biometric identification based
on ECG signal achieving accuracies of 94.9% [18] and 97.6% [23]. The most
common methods for ECG identification have been wavelet transform for fea-
ture extraction and Support Vector Machine for classification [6,10,17]. However,
there are no studies that define the effects of diseases or conditions in biometric
identification. The ECG signals have high power for the diagnosis of multiple
cardiac alterations. Therefore, some studies of biometric systems based on ECG
have considered alterations to improve its performance and reliability [21]. How-
ever, some heart diseases that cannot be detected throughout ECG signals affect
their morphology such as cardiac murmurs, which are considered as one of the
412 M. A. Becerra et al.
most common failures of the heart and they have been widely studied from
phonocardiograms and echocardiograms for detecting different pathologies [2,9].
Therefore, the primary objective of this work is to analyze the effects of car-
diac murmurs on biometric identification based on ECG signals which still have
not been studied (taking into account our review) The second objective is to
achieve an effective mechanism based on signal processing, and pattern recog-
nition techniques. In this study are studied 40 subjects without distinguishing
gender and age for human identification from ECG signals. 20 subject with-
out heart pathologies and 20 with cardiac murmurs. A database was collected
with the help of specialists in cardiology 8 continuous registers were acquired by
subject. Then, the signals were filtered, standardized, and manually segmented
by beat in the preprocessing stage. In the feature extraction step, the signals
were decompose using discrete wavelet transform, and maximal overlap discrete
wavelet transform, and different features were calculated using linear and non-
linear statistical measures alongside Mel frequency cepstral coefficients extracted
from the ECG signals. A relevant analysis was carried out using the Relief F algo-
rithm. Finally, five classifiers and a mixture of them were tested. The best global
result 91.19% was achieved using the LDC classifier.
2 Experimental Setup
This study was carried out in 5 main stages as it is shown in Fig.1. First, was col-
lected a database of ECG signals, then, segmentation and standardization were
applied in pre-processing step. In third stage feature extraction was carried out
using multiple techniques based on Mel Frequency cepstral coefficients (MFCC),
linear and non-linear measures of ECG signals, DWT, and MODWT. To reduce
the dimensionality of the feature space was performed relief F algorithm. Finally,
in the stage 5 was applied multiple classifiers as follows: Support Vector Machine
(SVM), Quadratic Bayes Normal Classifier (QDC), Optimisation of the Parzen
classifier (PZ), k-Nearest Neighbor (K-NN), and Linear Bayes Normal Classifier
(LDC) alongside a mixture of them [13].
Fig. 1. Experimental procedure
Exploratory Study of the Effects of Cardiac Murmurs 413
2.1 Database
320 ECG recordings correspond to 40 subjects (20 healthy subjects and 20 sub-
jects with heart disease) were selected from a database of ECG recordings col-
lected from 143 adult subjects of whom 55 patients were labeled as normal,
and 88 had evidence of cardiac murmurs (aortic stenosis, mitral regurgitation,
among others). The subjects gave their formal consent and underwent a medical
examination with the approval of the ethical committee. Cardiologists evalu-
ated the valve lesion severity according to a clinical routine. Eight recordings
of lasts 8 s were recorded from each subject in the phase of post-expiratory and
post-inspiratory apnea. The signals were acquired at 44.1 kHz with 16-bits per
sample. For this study were selected 40 subjects, 20 healthy and 20 with cardiac
murmurs. In Fig. 2are shown three normal ECG signals, and in Fig. 3are shown
three pathological signals, which evidence morphological differences generated
by cardiac murmurs.
Fig. 2. Normal ECG signals
Fig. 3. Pathological ECG signals
2.2 Preprocessing
The ECG signals were manually segmented per beat. Then a standardization
was applied to the amplitude of the signals between [−1 1]. Finally, to eliminate
baseline noise a Butterworth bandpass filter of second order was used with a
cutoff frequency of 0.5 Hz and cutoff 150 Hz.
414 M. A. Becerra et al.
2.3 Feature Extraction
The following techniques were applied to ECG segmented signals: (i)39MFCC
were calculated from ECG signals, including delta coefficients, delta–delta coef-
ficients, and log energy, using 24 Hamming shaped filters and sliding hamming
windows (50% overlap). (ii ) 65 features obtained of measures shown in Table 1
applied to coefficients of DWT db10 of 4 levels. (iii) 65 features obtained of mea-
sures shown in Table 1applied to coefficients of MODWT [1] db10 of 4 levels.
(iv ) 10 features obtained of measures shown in Table 1applied to ECG signals.
In total 179 features were extracted for each signal.
Table 1. Measures calculated from ECG signals - MODWT and DWT
Renyi entropy Root mean square Standard deviation Entropy Varia nce Power Energy
Shannon entropy Log energy entropy Covariance Kurtosis Min/max Mean
2.4 Feature Selection
The ReliefF algorithm is a filter method that ponders each feature according to
its relevance to each class. The weights of them are updated iteratively [24]. This
method was applied to 3 datasets for reducing the dimensionality of the following
feature space: (i) Features obtained from segmented pathological signals, (ii )
Features obtained from segmented normal signals, and (iii ) Features obtained
from segmented normal and pathological signals. The selection criteria was a
ranked cumulative above of 95%.
2.5 Classification
Support Vector Machine, QDC Bayesian classification, PARZEN, k-NN and LDC
classifiers, and a mixture among them were tested. Six methods of the mixture
were applied as follows: Product, Mean combiner, Median combiner, Maximum
combiner, Minimum combiner, and majority vote (Vm). These techniques are
explained in [13]. These classifiers were validated using cross-validation with
10-fold.
3 Results and Discussion
In Table 2are shown the results obtained by the different classifiers (SVM, QDC,
K-NN, LDC) and mixture of all them using the product, mean, median, max, min
and Vm operations. These classifiers systems were tested with the normal signals
(NS) dataset and pathological signals (PS) dataset and All (PS and NS datasets).
The best individual results 95.31% were obtained by PZ and K-NN classifiers for
PS. LDQ classifier achieved the best performance 97.74% of accuracy for Normal
Exploratory Study of the Effects of Cardiac Murmurs 415
Table 2. Accuracy of classifiers systems
Signals SVM QDC PZ K-NN LDC Prod Mean Median Max Min Vm
PS 71.09 81.25 95.31 95.31 92.97 85.94 90.62 89.84 90.62 79.69 94.53
NS 45.86 84.21 87.22 86.47 97.74 87.22 86.47 87.22 86.47 54.89 93.23
ALL 29.12 78.54 77.39 79.31 91.19 80.84 81.23 75.48 81.23 47.51 87.36
signals and 91.19 for all signals (PS and NS). The best results using a mixture
of classifier were achieved using Vm technique for mixturing.
This work is compared in Table 3with other studies of the biometric systems
based on ECG signals using the accuracy. The best global result presented in
this work is not comparable with the different approaches due to that the ECG
biometric studies do not analyze the effects of the cardiac murmurs, despite some
of them included into their studies normal and pathological ECG signals. The
best result 97.74% was achieved in this work using normal signals. Nonetheless,
the results obtained when were included ECG signals affected by cardiac mur-
murs were decreased 6.55%, resulting in performance lower than the showed by
SVM and ANN approaches.
Table 3. Comparison with other approaches
Approach SVM [23]ANN [23]SVM-OAA [7]LDC (this work-ALL) LDC (this work-NS)
Accuracy 96.6% 97.6% 88.41% 91.19% 97.74%
4 Conclusions
In this paper, a study of cardiac murmur effects on biometric identification based
on ECG signals was presented. Five classifiers and six mixture of them were
tested using normal and pathological signals individually and together. LDQ
classifier shown the best global performance 91.19% and separately for normal
signals 97.74%. K-NN and PZ classifiers shown the best results 95.31% for patho-
logical signals. The majority vote technique for the mixture of classifiers shows
the best results, but these were lower compared with the performance of indi-
vidual classifiers.
Based on the results achieved with the pathological ECG signals vs. nor-
mal ECG signals for human identification was demonstrated that the cardiac
murmurs affect the biometric identification based on ECG signals and generate
opposite effects on identification performance between PS and NS. This effect
can be explained, taking into account that cardiac murmurs are caused when the
blood flow becomes turbulent near damaged cardiac valves that can have low
generalization and they generate electrical changes in the cardiac muscle which
elicits changes in the ECG signals. In conclusion, the cardiac murmur effects on
the biometric system based on ECG signals is a characteristic that affects the
416 M. A. Becerra et al.
performance of the system, but no helps to human identification although the
accuracy of the system can be upper for some classifiers using only pathological
signals. Nevertheless, the global accuracy of identification including pathological
and normal signals decrease the performance of the system. As future work, we
proposed to analyze effects of post-expiratory and post-inspiratory apnea phases
when acquiring ECG signals for biometric systems, and include another type of
cardiac diseases into the study to achieve a significant generality of biometric
systems based on ECG signals.
Acknowledgment. The authors acknowledge to the research project “Desarrollo de
una metodolog´ıa de visualizaci´on interactiva y eficaz de informaci´on en Big Data”
supported by Agreement No. 180 November 1st, 2016 by VIPRI from Universidad
de Nari˜no. As well, authors thank the valuable support given by the SDAS Research
Group (www.sdas-group.com).
References
1. Barzegar, R., Asghari Moghaddam, A., Adamowski, J., Ozga-Zielinski, B.: Multi-
step water quality forecasting using a boosting ensemble multi-wavelet extreme
learning machine model. Stoch. Environ. Res. Risk Assess. 32(3), 799–813 (2018).
https://doi.org/10.1007/s00477-017-1394- z
2. Becerra, M.A., Orrego, D.A., Mejia, C., Delgado-Trejos, E.: Stochastic analysis and
classification of 4-area cardiac auscultation signals using Empirical Mode Decompo-
sition and acoustic features. Comput. Cardiol. (CinC) 2012(September), 529–532
(2012)
3. Da Silva Luz, E.J., Moreira, G.J., Oliveira, L.S., Schwartz, W.R., Menotti, D.:
Learning deep off-the-person heart biometrics representations. IEEE Trans. Inf.
Forensics Secur. 13(5), 1258–1270 (2018). https://doi.org/10.1109/TIFS.2017.
2784362
4. Dar, M.N., Akram, M.U., Shaukat, A., Khan, M.A.: ECG based biometric iden-
tification for population with normal and cardiac anomalies using hybrid HRV
and DWT features. In: 2015 5th International Conference on IT Convergence
and Security, ICITCS 2015 - Proceedings (2015). https://doi.org/10.1109/ICITCS.
2015.7292977
5. Elhoseny, M., Essa, E., Elkhateb, A., Hassanien, A.E., Hamad, A.: Cascade mul-
timodal biometric system using fingerprint and iris patterns. In: Hassanien, A.E.,
Shaalan, K., Gaber, T., Tolba, M.F. (eds.) AISI 2017. AISC, vol. 639, pp. 590–599.
Springer, Cham (2018). https://doi.org/10.1007/978-3-319-64861-3 55
6. Hejazi, M., Al-Haddad, S.A., Hashim, S.J., Aziz, A.F.A., Singh, Y.P.: Feature
level fusion for biometric verification with two-lead ECG signals. In: Proceeding -
2016 IEEE 12th International Colloquium on Signal Processing and its Applica-
tions, CSPA 2016, March, pp. 54–59 (2016). https://doi.org/10.1109/CSPA.2016.
7515803
7. Hejazi, M., Al-Haddad, S.A., Singh, Y.P., Hashim, S.J., Abdul Aziz, A.F.: ECG
biometric authentication based on non-fiducial approach using kernel methods.
Digit. Signal Process.: Rev. J. 52, 72–86 (2016). https://doi.org/10.1016/j.dsp.
2016.02.008
Exploratory Study of the Effects of Cardiac Murmurs 417
8. da Silva, H.P., Carreiras, C., Lourenco, A., Fred, A., das Neves, R.C., Ferreira, R.:
Off-the-person electrocardiography: performance assessment and clinical correla-
tion. Health Technol. 4, 309–318 (2015)
9. Jimenez, J., Becerra, M., Delgado-Trejos, E.: Heart murmur detection using ensem-
ble empirical mode decomposition and derivations of the mel-frequency cepstral
coefficients on 4-area phonocardiographic signals. In: Computing in Cardiology,
vol. 41 (2014)
10. Jung, W.H., Lee, S.G.: ECG identification based on non-fiducial feature extraction
using window removal method. Appl. Sci. 7(12), 1205 (2017). https://doi.org/10.
3390/app7111205.http://www.mdpi.com/2076-3417/7/11/1205
11. Kanchan, T., Krishan, K.: Loss of fingerprints: forensic implications. Egypt. J.
Forensic Sci. 8(1), 19 (2018). https://doi.org/10.1186/s41935-018-0051-0
12. Martinez-Diaz, M., Fierrez, J., Galbally, J., Ortega-Garcia, J.: An evaluation of
indirect attacks and countermeasures in fingerprint verification systems. Pattern
Recognit. Lett. 32(12), 1643–1651 (2011). https://doi.org/10.1016/J.PATREC.
2011.04.005
13. Moreno-Revelo, M., Ortega-Adarme, M., Peluffo-Ordo˜nez, D.H., Alvarez-Uribe,
K.C., Becerra, M.A.: Comparison among physiological signals for biometric iden-
tification. In: Yin, H., et al. (eds.) IDEAL 2017. LNCS, vol. 10585, pp. 436–443.
Springer, Cham (2017). https://doi.org/10.1007/978-3-319-68935-7 47
14. Murillo-Escobar, M., Cruz-Hern´andez, C., Abundiz-P´erez, F., L´opez-Guti´errez,
R.: A robust embedded biometric authentication system based on fingerprint and
chaotic encryption. Expert. Syst. Appl. 42(21), 8198–8211 (2015). https://doi.org/
10.1016/j.eswa.2015.06.035
15. Odinaka, I., Lai, P.H., Kaplan, A.D., O’Sullivan, J.A., Sirevaag, E.J., Rohrbaugh,
J.W.: ECG biometric recognition: a comparative analysis (2012). https://doi.org/
10.1109/TIFS.2012.2215324
16. Pal, S., Mitra, M.: Increasing the accuracy of ECG based biometric analysis by
data modelling. Measurement 45(7), 1927–1932 (2012). https://doi.org/10.1016/
J.MEASUREMENT.2012.03.005
17. Patro, K., Kumar, P.: Machine learning classification approaches for biometric
recognition system using ECG signals. J. Eng. Sci. Technol. Rev. 10(6), 1–8 (2017).
https://doi.org/10.25103/jestr.106.01
18. Pinto, J., Cardoso, J., Louren¸co, A., Carreiras, C.: Towards a contin-
uous biometric system based on ECG signals acquired on the steer-
ing wheel. Sensors 17(10), 2228 (2017). https://doi.org/10.3390/s17102228.
http://www.mdpi.com/1424-8220/17/10/2228
19. Sidek, K.A., Khalil, I., Jelinek, H.F.: ECG biometric with abnormal cardiac condi-
tions in remote monitoring system. IEEE Trans. Syst., Man, Cybern.: Syst. 44(11),
1498–1509 (2014). https://doi.org/10.1109/TSMC.2014.2336842
20. Song, W., Kim, T., Kim, H.C., Choi, J.H., Kong, H.J., Lee, S.R.: A finger-vein
verification system using mean curvature. Pattern Recognit. Lett. 32(11), 1541–
1547 (2011)
21. Tan, R., Perkowski, M.: Toward improving electrocardiogram (ECG) biometric
verification using mobile sensors: a two-stage classifier approach. Sensors 17(2),
410 (2017)
22. Zapata, J.C., Duque, C.M., Rojas-Idarraga, Y., Gonzalez, M.E., Guzm´an, J.A.,
Becerra Botero, M.A.: Data fusion applied to biometric identification – a review.
In: Solano, A., Ordo˜nez, H. (eds.) CCC 2017. CCIS, vol. 735, pp. 721–733. Springer,
Cham (2017). https://doi.org/10.1007/978-3-319- 66562-7 51
418 M. A. Becerra et al.
23. Zhang, Y., Wu, J.: Practical human authentication method based on piecewise
corrected Electrocardiogram. In: Proceedings of the IEEE International Conference
on Software Engineering and Service Sciences, ICSESS (61571268), pp. 300–303
(2017). https://doi.org/10.1109/ICSESS.2016.7883071
24. Zhang, Z., Li, Y., Jin, S., Zhang, Z., Wang, H., Qi, L., Zhou, R.: Modulation signal
recognition based on information entropy and ensemble learning. Entropy 20(3),
198 (2018)