Professional Documents
Culture Documents
Reconstruction of Depolarization Pattern in Myocar
Reconstruction of Depolarization Pattern in Myocar
Reconstruction of Depolarization Pattern in Myocar
Abstract. Our paper introduces a new technique for diagnosis of various heart diseases
without the need of highly experts to investigate the electrocardiogram (ECG). Using
the same electrodes of the ECG machine, it will be able to transmit directly the electrical
activity inside the heart to a moving picture. Our technique is based on artificial
intelligence algorithm using artificial neural networks (ANN). Finding the trans-
membrane potential (TMP) inside the heart from the body surface potential (BSP) is
known as the inverse problem of ECG. To have a unique solution for the inverse
problem the data used should be obtained from a forward model. A three dimensional
(3-D) model of cellular activation whole heart embedded in torso is simulated and
solved using COMSOL Multiphysics software. In our previous paper, one ANN
succeeded in displaying the wave propagation on the surface of a normal heart. In this
paper, we used a configuration of ANNs to display different cases of heart with
myocardial infarction (MI). To check the system accuracy, eight MI cases with different
sizes and locations in the heart are simulated in the forward model. This configuration
proved to be highly accurate in displaying each MI case -size and location- presenting
the infarction as an area with no electrical activity.
1. Introduction
The MI results from death of the heart muscle due to the blockage in the coronary artery. However, the
localization of MI mainly depends on the visual analysis of ECG [1-3], which strongly depends on the
physician experience. Therefore, it is very important to find a reliable method to visualize MIs directly
inside the heart. Reconstruction of MI by solving the inverse problem of ECG is a non-invasive and
relative low cost method to localize MI accurately. Recently, many methods have been introduced for
improving the reconstruction quality of MI by solving the inverse problem of ECG using regularization
techniques [4-7].
The inverse problem of ECG, also called the cardiac electrical source imaging, is to reconstruct the
electrical activities in the heart (TMP) from multichannel ECG measured on the body surface (BSP) [4,
8-13]. Also it should be noted that the transfer function between the TMP and the BSP is in fact highly
nonlinear. Different papers used ANN to solve mathematical inverse problems [14-19].
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ICaTAS 2021 IOP Publishing
Journal of Physics: Conference Series 2128 (2021) 012016 doi:10.1088/1742-6596/2128/1/012016
Most of the work done to detect and localize MI uses the visual analysis of ECG [1-3]. Other studies
reconstruct the MI by solving the inverse problem of ECG using regularization techniques [4-6]. Most
of these work focuses only on localizing the MI size and position in the heart. In our previous work, we
used one ANN to reconstruct the depolarization pattern for normal heart and for one MI case [20]. But,
with the increase in number of MI cases, the localization performance of an ANN classifier goes poor.
This paper focuses on visualizing the electrical activity inside the heart as a real time moving picture
during the depolarization period of 100ms for normal heart and different cases of MIs. In this paper, we
used two cascaded stages of ANNs to solve the inverse problem of ECG by reconstructing the
depolarization pattern (TMP) for different MI cases. Each case could be localized in the heart with its
accurate size in a minimal processing time. To check the system accuracy, eight MI cases of different
sizes and at various sites in the ventricles are simulated as eight different forward problems. Each case
was simulated by real heart equations (FitzHugh-Nagumo equations) using COMSOL Multiphysics
finite element software [9-11]. For the learning process of ANNs, TMP is computed for 74 different
positions covering the whole heart; BSP is computed for 9×9 array of electrodes placed on the torso
surface opposite to the heart. ANN is trained by the TMP and the BSP values computed all over the
depolarization period of 100 ms within the cardiac cycle. For the new configuration of two cascaded
stages of ANNs, the primary ANN is responsible for selecting the MI case and localizing the infarction
position and size in the heart. However, the secondary ANN is responsible for reconstructing the
depolarization pattern for the selected MI case. This configuration is also examined by two small
partially overlapped MI regions which are difficult to be detected and localized.
The rest of this paper is classified as shown. Section 2 illustrates the forward and inverse models used
in details including the heart model and the new proposed configuration of ANNs. The simulation results
of the forward and the inverse methods are displayed and discussed in section 3. Section 4 is devoted to
the main conclusions.
2. Model analysis
The TMP is measured at 74 different positions covering the surface of the heart as shown in Figure 1(a).
The BSP is measured at 81 different positions distributed as 9×9 array on the torso surface opposing the
heart as shown in Figure 1 (b). These values are measured 100 successive times for a depolarization
period of 100 ms.
(a) (b)
Figure 1. (a) The 74 TMP positions distributed on the heart surface and (b) the 81 (9×9) BSP positions
on the torso surface.
2
ICaTAS 2021 IOP Publishing
Journal of Physics: Conference Series 2128 (2021) 012016 doi:10.1088/1742-6596/2128/1/012016
In this study eight different MI cases are simulated in the forward model and reconstructed in the inverse
model. For each MI case a new forward model is simulated with a different infarcted region of different
size and location in the ventricles. Figure 2 shows three different MI cases, the infarction in each case is
indicated by an area with no activation presented in blue color. Figure 2(a) illustrates the infarction
location in the heart in the XY plan; the infarction size is represented in Figure2(b). To check our system
accuracy for more difficult MI cases, we simulated two smaller partially overlapped MI regions (the last
case shown in Figure 2). All infarctions are positioned spanning the epicardium, midmyocardium and
endocardium. For each MI case, the TMP is measured at the 74 positions covering the heart shown in
Figure1(a) and the BSP is measured at the 81 positions on the torso surface shown in Figure 1(b). For
each case the TMP and the BSP values are measured 100 successive times for a one complete
depolarization period.
(a) (b)
Figure 2. Three different MI cases, (a) The infarction location in the heart in the XY plan and (b)
represents the infarction size.
3
ICaTAS 2021 IOP Publishing
Journal of Physics: Conference Series 2128 (2021) 012016 doi:10.1088/1742-6596/2128/1/012016
Figure 3.The new configuration of two cascaded stages of ANNs block diagram.
The primary ANN is trainedby the TMP and the BSP values of the normal heart and all the MI cases
during the depolarization of the ventricles only. Its input is theBSP at a certain millisecond for one of
the nine heart cases shown in Figure3. The primary ANN role is to detect the heart case from the given
input according to the TMP and the BSP values learnt. The selection program activates the secondary
ANN according to the detected heart case. Each secondary ANN is trained by TMP and BSP for each
single heart case during one complete depolarization period. The output of the secondary ANN is the
TMP corresponding to the input BSP at a certain millisecond. For one complete depolarization period,
each input BSP from the outer surface electrodes gives an output of its corresponding TMP obtaining a
moving picture for the electrical activity inside the heart.
4
ICaTAS 2021 IOP Publishing
Journal of Physics: Conference Series 2128 (2021) 012016 doi:10.1088/1742-6596/2128/1/012016
FORWARD INVERSE
(a) (b)
Figure 4.Three selected MI cases (a) in the forward model, (b) in the inverse model.
5
ICaTAS 2021 IOP Publishing
Journal of Physics: Conference Series 2128 (2021) 012016 doi:10.1088/1742-6596/2128/1/012016
(a)
Inverse
(b)
Figure 5.𝑉𝑚 (TMP) of the MI region 7 at four different milliseconds (a) in the forward method, (b) in
the inverse method of two cascaded stages of ANNs
The new proposed configuration of ANNs was examined for the eight different MI cases presented in
this paper. It succeeded toreconstruct the depolarization pattern for each case includingthe two smaller
partially overlapped cases.
4. Conclusion
Using two cascaded stages of ANNs shows a great success inreconstructing the depolarization pattern
fornormal heart and different MI cases from the BSPs computed in the forward model. The new
configuration gives an accurate estimation for the infarctions sizes and positions in the heart for eight
MI case, including difficult cases,with almost zero error compared to the forward equivalent constructed
cases.Furthermore, this technique can dispense with the complicated investigation of ECGgiving
directly a complete image of the electrical activity inside the heart during the depolarization period.
References
[1] Arif M, Malagore I A and Afsar F A 2012 Detection and localization of myocardial infarction using
K-nearest neighbor classifier J. Medical Systems 36 279-289
[2] Arif M, Malagore I A and Afsar F A 2010 Automatic detection and localization of myocardial
infarction using back propagation neural networks Proceedings of 4th IEEE International Conference
on Bioinformatics and Biomedical Engineering (iCBBE), Chengdu, China 1-4
[3] Ravish D, Shenoy N, Shanthi K and Nisargh S 2014 Heart function monitoring prediction and
prevention of heart attacks: Using artificial neural networks Proceedings of the International
Conference on Contemporary Computing and Informatics (IC3I), Mysore, India 1-6
[4] Chavez C E, Zemzemi N, Coudière Y, Alonso-Atienza F, Chavez C E, Zemzemi N and Alvarez D
Inverse problem of electrocardiography: estimating the location of cardiac isquemia in a 3D
geometry Functional Imaging and Modelling of the Heart (FIMH2015), Maastricht, Netherlands.
Springer International Publishing 393-401
[5] Yao B, Zhu R and Yang H 2017 Characterizing the location and extent of myocardial infarctions with
inverse ECG modeling and spatiotemporal regularization IEEE Journal of Biomedical and Health
Informatics 22 1445-1455
6
ICaTAS 2021 IOP Publishing
Journal of Physics: Conference Series 2128 (2021) 012016 doi:10.1088/1742-6596/2128/1/012016
[6] Jiang Y, Qian C, Hanna R, Farina D and Dossel O 2009 Optimization of the electrode positions of
multichannel ECG for the reconstruction of ischemic areas by solving the inverse
electrocardiographic problem International Journal of Bioelectromagnetism 11 27-37
[7] Li G and He B 2004 Noninvasive estimation of myocardial infarction by means of a heart-model-
based imaging approach—a simulation study. Med Biol. Eng. Comput. 42 128-136
[8] Jiang M, Jiang S, Zhu L, Wang Y, Huang W and Zhang H 2013 Study on parameter optimization for
support vector regression in solving the inverse ECG problem Computational and Mathematical
Methods in Medicine 2013 9 pages
[9] Sovilj S, Magjarevi´c R, Lovell N H and Dokos S 2013 Simplified 3D model of whole heart electrical
activity and 12- lead ECG generation Computational and Mathematical Methods in Medicine 2013
10 pages
[10] Dokos S, Cloherty S L and Lovell N H 2007 Computational model of atrial electrical activation and
propagation Proceedings of the 29th Annual International Conference of the IEEE (EMBS), Lyon,
France 908-911
[11] Sovilj S, Čeperičand V and Magjarevič P R 2016 3D cardiac electrical activity model Automatika
57 540-548
[12] Shou G, Xia L, Jiang M, Liu F and Crozier S 2007 Forward and inverse solutions of
electrocardiography problem using an adaptive BEM method Proceedings of the 4th International
Conference on Functional Imaging and Modeling of the Heart(FIMH’07), Salt Lake City, UT, USA
290-299
[13] Jiang M, Xia L, Shou G, Wei Q, Liu F and Crozier S 2009 Effect of cardiac motion on solution of
the electrocardiography inverse problem IEEE Transactions on Biomedical Engineering 56 923-931
[14] Khalifa A M 1997 The inverse problem in electrocardiography by using the Fourier transform of the
body surface potential: a study on a homogeneous spherical model In Signal Analysis and Prediction
I, EURASIP, ICT Press, Prague, Czech Republic 391-394
[15] Khalifa A M 1997 The use of neural-network for detection of multi-dipoles in a homogenous
spherical conductor: a proposed technique for the inverse problem of electrocardiography. In Signal
Analysis and Prediction I, EURASIP, ICT Press, Prague, Czech Republic 395-398
[16] Khalifa A M, AbdelRazek M and Nada S 1997 The ANN as a nonlinear regularization technique to
solve the inverse problem of electrocardiography Proceedings of the IEEE Engineering in Medicine
and Biology Society Region 8 International Conference, Prague, Czech Republic 68-69
[17] Gorbachenko V I, Lazovskaya T V, Tarkhov D A, Vasilyev A N and Zhukov M V 2016 Neural
network technique in some inverse problems of mathematical physics Proceedings of the 13th
International Symposium on Neural Networks 2016 (ISNN), St. Petersburg, Russia, 310-316
[18] Jin K H, McCann M T, Froustey E and Unser M 2017 Deep convolutional neural network for inverse
problems in imaging IEEE Trans. Image Process 26 4509-4522
[19] Adler J and Öktem O 2017 Solving ill-posed inverse problems using iterative deep neural networks.
Inverse Problems 33 1-24
[20] Mabrouk N A, Khalifa A M, Nasser A A and Aly M H 2018 Reconstructing the heart depolarization
pattern from body surface potentials using artificial neural network The 3rdInternational Conference
on Advanced Technology and Applied Sciences (ICaTAS), Sepang, Malaysia 1570464928