1 Patient-Specific Heart Model Towards Atrial Fibrillation

2025-04-30 0 0 6.98MB 11 页 10玖币
侵权投诉
1
Patient-Specific Heart Model Towards Atrial
Fibrillation
Jiyue He, Arkady Pertsov, Sanjay Dixit, Katie Walsh, Eric Toolan, Rahul Mangharam
Abstract—Atrial fibrillation is a heart rhythm disorder that
affects tens of millions people worldwide. The most effective
treatment is catheter ablation. This involves irreversible heat-
ing of abnormal cardiac tissue facilitated by electroanatomical
mapping. However, it is difficult to consistently identify the
triggers and sources that may initiate or perpetuate atrial
fibrillation due to its chaotic behavior. We developed a patient-
specific computational heart model that can accurately reproduce
the activation patterns to help in localizing these triggers and
sources. Our model has high spatial resolution, with whole-
atrium temporal synchronous activity, and has patient-specific
accurate electrophysiological activation patterns. A total of 15
patients data were processed: 8 in sinus rhythm, 6 in atrial
flutter and 1 in atrial tachycardia. For resolution, the average
simulation geometry voxel is a cube of 2.47 mm length. For
synchrony, the model takes in about 1,500 local electrogram
recordings, optimally fits parameters to the individual’s atrium
geometry and then generates whole-atrium activation patterns.
For accuracy, the average local activation time error is 5.47 ms
for sinus rhythm, 10.97 ms for flutter and tachycardia; and the
average correlation is 0.95 for sinus rhythm, 0.81 for flutter and
tachycardia. This promising result demonstrates our model is
an effective building block in capturing more complex rhythms
such as atrial fibrillation to guide physicians for effective ablation
therapy.
Index Terms—medical cyber-physical systems, patient-specific,
computational heart model, cardiac electrophysiology modeling,
electroanatomical mapping, Mitchell-Schaeffer model
I. INTRODUCTION
Atrial Fibrillation (AF) is a heart rhythm disorder where
the normal beating in the two atria is irregular, and blood
does not flow well to the ventricles. While the underlying
mechanism of AF is not clearly understood, it largely is due
to the spatial variation in the conduction properties of the atrial
myocardium. This causes the single wavefront propagating
across the atria to being split into multiple wavefronts resulting
in chaotic depolarization of the heart tissue and irregularly fast
rhythm. AF increases the risk of stroke and heart failure. If
left untreated, it will become worse [1].
One of the most effective treatments for AF is catheter
ablation. This involves irreversible radiofrequency heating of
the sources that initiate or perpetuate AF. The common current
ablation protocol involves standard lesion locations for all
patients. For example, in the Pulmonary Vein Isolation (PVI)
Jiyue He, Rahul Mangharam: Department of Electrical and Sys-
tems Engineering, University of Pennsylvania, Philadelphia, PA, USA.
jiyuehe@seas.upenn.edu. Arkady Pertsov: Department of Pharmacology, Up-
state Medical University, Syracuse, USA. Sanjay Dixit, Katie Walsh: Depart-
ment of Cardiac Electrophysiology, Hospital of the University of Pennsylva-
nia, Philadelphia, PA, USA. Eric Toolan: Biosense Webster, Lansdowne, PA,
USA. May 19, 2021.
approach, ablation lesions encircle the pulmonary veins to
prevent abnormal activations originated in the veins travel into
the atria.
Another common protocol involves PVI with additional ab-
lation of non-pulmonary vein triggers and putative arrhythmia
substrate [2]. While paroxysmal-AF (i.e. spontaneous onset
and termination) can be eliminated in 70-75% of patients with
a single ablation procedure, persistent-AF can be eliminated
in only about 50% of patients with a single procedure [3].
The reason is there are triggers other than pulmonary veins
that cause fibrillation for the persistent-AF. Electroanatomical
mapping captures the heart geometry and tissue conductivity
which are essential in identifying those trigger [4].
The central challenge is to capture high-resolution synchro-
nized electrograms across the entire atrium and analyze these
data to identify potential triggers as ablation candidates to ter-
minate AF. We developed a high spatial resolution, temporally
synchronous and patient-specific atrium model which captures
electrophysiologic and anatomic parameters unique to each
patient.
Figure 1 shows the overall process of constructing the
heart model. First, a high resolution electroanatomical map
is exported from the Carto3 System, which contains atrium
3D triangular mesh, electrograms, and electrode locations.
The mesh is processed to remove geometry defects, such as
deep concave holes, intersecting triangular faces, and non-
referenced vertices. The mitral valve and 4 pulmonary veins
are cut out. Then a 3D Cartesian grid is generated and wrapped
around the mesh for computing simulation. Then, these are put
into the electrophysiological heart model, and patient-specific
parameters are fitted via an optimization process. As a result
of fitting parameters at every locations on the mesh, the heart
model becomes whole-atrium synchronous. Lastly, the heart
model is validated over 15 patients data by comparing Local
Activation Time (LAT) between patient data and simulation
data.
This model demonstrates the ability to accurately produce
the activation patterns for an individual patient that can better
identify AF sources.
II. RELATED WORK
Several research groups had developed patient specific high
resolution computational heart models [5]. Cabrera-Lozoya et
al developed a model constructed from 3D Delayed-Enhanced
Magnetic Resonance Imaging (DE-MRI) to simulate patient-
specific post-infarction Ventricular Tachycardia (VT) abnor-
mal electrograms. From DE-MRI data, they segmented the
arXiv:2210.12825v1 [physics.med-ph] 23 Oct 2022
2
Fig. 1. The overall process. Patient data consists of electrode locations, electrograms, and atrium 3D triangular mesh. Mesh is refined and pulmonary veins
and mitral valve are cut out. Then, a Cartesian grid is created wrapping around the mesh. These data are input into the electrophysiological heart model,
where parameters are fitted via an optimization process. Lastly, the heart model is validated through patient data.
ventricle into healthy myocardium, scar, and border zone.
Then, three different sets of parameters were assigned to
these three regions. They showed that their model can gen-
erate distinguishable electrograms for healthy region and scar
region [6]. Boyle et al developed a model for personalized
ablation guidance of AF. They obtained atrium geometry
and fibrosis distribution from Late Gadolinium Enhancement
Magnetic Resonance Imaging (LGE-MRI). Healthy regions
and fibrosis regions are assigned with different parameters.
Then virtual pacing are performed in simulation to identify
all possible ablation targets [7]. Lim et al developed a model
to do simulation-guided catheter ablation of AF. Their model
combined data from Computed Tomography (CT) images and
electroanatomical maps to capture anatomy, fiber orientation,
fibrosis, and electrophysiology. Different set of conduction
parameters are assigned to fibrotic and non-fibrotic tissue [8].
Corrado et al developed a model that had locally fitted param-
eters. They applied an S1-S2 electrical stimulation protocol
from the coronary sinus and the high right atrium and recorded
endocardium electrograms in the left atrium. The parameter
fitting was done via a grid-search method that evaluated all
the combinations of parameters within a range [9] [10].
The major differences of our heart model are:
Our model identifies tissue conduction and diffusion
values locally at the voxel level, while other heart models
assign two or three sets of fixed parameters to generic
healthy or scar regions.
Our model fits the local parameters using self-activated
electrograms rather then manual pacing-induced ones.
Our model does not use MRI or CT because they do not
provide electrical activation data. Instead, we use elec-
troanatomical mapping data which provides endocardium
electrogram.
Our model is computationally light and will eventually
be able to run during the clinical procedure for real-time
ablation guidance.
It takes about 1 to 2 hours for Carto3 System to export 1
patient data. Our heart model program is implemented using
Matlab (MathWorks), and runs on a laptop with 1 Intel Core i7
CPU. It takes about 2 minutes to read in a patient data to the
computer. On average, it takes about 50 seconds to personalize
one heart model. If our heart model was integrated into the
Carto3 System, the step of export/import data would not be
needed, which means the entire computation could be finished
in 50 seconds. Further more, it could speed up hundreds times
if GPU computing was implemented.
III. HIGH SPATIAL RESOLUTION
A. Data Collection
At the start of the catheter ablation procedure, the physician
captures an electroanatomical map with a roving mapping
catheter from the Carto3 System at the Hospital of the Uni-
versity of Pennsylvania. As show in Figure 2, it associates
the 3D atrium triangular mesh with electrical activity at every
location. The mesh mapping fill threshold was 5 mm, and
electrogram recording filters were set at 2 to 240 Hz for
unipolar electrograms, 16-500 Hz for bipolar electrograms,
and 0.5-200 Hz for surface electrogram. Each map has about
1,500 electrogram recordings spread across the endocardium.
Each electrogram records 2.5 seconds unipolar and bipolar
signals at 1 kHz. Bad electrograms will introduce noise to the
model, thus needed to be excluded:
Electrode distance to the nearest mesh vertex is greater
than 8 mm. This is an empirical threshold we decided to
3
Fig. 2. (a) Pentaray catheter is a star-shape catheter that has 20 electrodes. (b) Physicians will hold the catheter in a location for 2.5 seconds to obtain one
segment of recording. (c) Then move the catheter to another location. (d) After 10 minutes, about 1,500 electrograms at different locations will be recorded.
use. The electrode was most likely not in contact with
atrium tissue beyond this threshold.
Maximum voltage is less than 0.45 mV. These electrodes
are either too far way from tissue or in contact with scar
tissue [11]. Neither will provide clean electrogram.
Complex and fractionated electrogram as shown in Figure
3. Fractionated electrograms consist of multiple high
frequency components with low amplitudes and long
duration, makes it difficult to find the accurate activation
time.
Fig. 3. (a) Electrogram is too complex and fractionated, making it difficult
to find the activation time. (b) Good electrogram. The activation time is easy
to identify.
The amount of raw electrogram recordings captured and fil-
tered are shown in Table I. The percentage of good electrogram
recordings is low for flutter and tachycardia maps because they
contain more fractionated electrograms.
B. Triangular Mesh to Cartesian Grid
To achieve accurate activation wave propagation, we cut out
the mitral valve and pulmonary veins. The cutting decision is
based on atrium anatomy [12]. The mesh is then refined to
make each of the triangles the same size and shape. Mesh
editing is done using a software called MeshLab [13], and the
two most helpful functionalities are 1) Simplification: quadric
edge collapse decimation, can be used to smooth mesh imper-
fections; and 2) Remeshing: isotropic explicit remeshing, can
be used to make triangles uniform in size and shape. For our
heart model to simulate atrium action potential propagation, it
requires a calculation of the second derivative of space. This
is difficult to perform on triangular mesh because the vertices
TABLE I
NUMBER OF ELECTRODE RECORDINGS
ID Rhythm # Electrode # Used % Used
1 Sinus Rhythm 976 557 57.1
2 Sinus Rhythm 3263 1361 41.7
3 Sinus Rhythm 3156 1788 56.7
4 Sinus Rhythm 1488 663 44.6
5 Sinus Rhythm 2477 1655 66.8
6 Sinus Rhythm 2905 2079 71.6
7 Sinus Rhythm 1744 861 49.4
8 Sinus Rhythm 1801 1106 61.4
Average 2226 1259 56.1
9 Flutter 278 73 26.3
10 Flutter 958 326 34.0
11 Tachycardia 822 197 24.0
12 Flutter 484 215 44.4
13 Flutter 1227 198 16.1
14 Flutter 714 130 18.2
15 Flutter 1469 425 28.9
Average 850 223 27.4
are not regularly positioned. In contrast, a Cartesian grid is
regular. In Figure 4, the black line represents a cross section
of the atrium and the blue dots are Cartesian grid voxels.
First, a Cartesian grid that encloses the entire atrium is created
as shown in (a). Then the voxels that are beyond a distance
threshold to the atrium mesh are removed as shown in (b).
The distance threshold is equal to 1.5 times the average inter
vertex distance. (c) (d) show the same process implemented
on a patient’s atrium mesh.
Patient electrograms are processed and the results are as-
signed to the nearest atrium mesh vertices. From atrium mesh
vertices, values will be projected onto the nearest Cartesian
grid voxels for simulation. Then the simulated results will
be projected back to the nearest vertices of the atrium mesh.
Denote dvoxel the distance in between two neighboring Carte-
sian grid voxels, dvertex the average distance in between two
neighboring atrium mesh vertices. It is important that this
equation to be satisfied: dvoxel <1/2×dvertex. Figure 5
explains why. The triangular mesh is the atrium, and the
larger dots are the Cartesian grid voxels. (a) If voxel spacing
is too large, when projecting values from vertices to voxels,
multiple vertices may be projected to the same voxel, causing
values to be overwritten and information lost. Here the two
red vertices are projected to the same green voxel. (b) If voxel
spacing is small enough, each red vertex will be projected to
摘要:

1Patient-SpecicHeartModelTowardsAtrialFibrillationJiyueHe,ArkadyPertsov,SanjayDixit,KatieWalsh,EricToolan,RahulMangharamAbstract—Atrialbrillationisaheartrhythmdisorderthataffectstensofmillionspeopleworldwide.Themosteffectivetreatmentiscatheterablation.Thisinvolvesirreversibleheat-ingofabnormalcard...

展开>> 收起<<
1 Patient-Specific Heart Model Towards Atrial Fibrillation.pdf

共11页,预览3页

还剩页未读, 继续阅读

声明:本站为文档C2C交易模式,即用户上传的文档直接被用户下载,本站只是中间服务平台,本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私,请立即通知玖贝云文库,我们立即给予删除!
分类:图书资源 价格:10玖币 属性:11 页 大小:6.98MB 格式:PDF 时间:2025-04-30

开通VIP享超值会员特权

  • 多端同步记录
  • 高速下载文档
  • 免费文档工具
  • 分享文档赚钱
  • 每日登录抽奖
  • 优质衍生服务
/ 11
客服
关注