G-Heart: A GPU-based System for
Electrophysiological Simulation and
Multi-modality Cardiac Visualization
Lei Zhang
a,b
, Kuanquan Wang
a
, Wangmeng Zuo
a
, Changqing Gai
a
a
School of Computer Science and Technology, Harbin Institute of Technology, Harbin 150001, China
Email: {wangkq, wmzuo}@hit.edu.cn
b
School of Art and Design, Harbin University, Harbin 150086, China
Email: cszhanglei@gmail.com
Abstract— Cardiac electrophysiological simulation and
multi-modality visualization are computationally intensive
and valuable in studying the structure, mechanism, and dy-
namics of heart. The existing multi-CPU based approaches
can reduce the calculation time, but suffer from the hard-
ware and communication cost problems and are inefficient
for 3D data visualization. Compared with multi-CPU, the
highly parallel and multi-core properties of GPU make it a
suitable alternative for accelerating cardiac simulation and
visualization. In this paper, we develop a G-Heart system
where GPU-based acceleration technologies are adopted for
both the simulation of cardiac electrophysiological activities
and the online illustration 3D multi-modality (anatomical
and electrophysiological) data. In the simulation stage, a
phase-field method is employed to cope with the no-flux
boundary condition. For heart geometrical structure illustra-
tion, a GPU-based ray-casting volume rendering algorithm
is implemented and an improved context-preserving model
with user interaction is integrated into the proposed frame-
work. Finally, a fusion visualization method is proposed,
which can provide 3D visualization results for both the
simulation data and the anatomical data simultaneously.
Index Terms— visualization, heart modelling, electrophysio-
logical simulation, CUDA, GPGPU
I. INTRODUCTION
C
Ardiac electrical activities are valuable for the in-
vestigation of complex heart diseases, e.g., arrhyth-
mias, ischemia, and ventricular fibrillation, and can be
invasively measured by medical devices like electro-
cardiogram (ECG). With the progress in programmed
electrical modeling and stimulation, electrophysiological
simulation has gradually been a promising direction for
cardiac electrophysiology study. By simulating the cardiac
electrical activities over a wide variety of scales from
single ion channel proteins to whole organs, cardiac
electrophysiological simulation can help in revealing the
mechanisms of normal and abnormal cardiac electrical
activities, interpreting of clinical data, and even designing
drugs and therapeutic plans [1], [2] .
Over the last decades, cardiac electrophysiological sim-
ulation has received considerable research interests, and
numerous models have been developed for simulating the
Corresponding author: Kuanquan Wang.
functions of protein, single cell, tissue, and whole heart
[3]. Recently, benefited by the progress in electrophys-
iological modeling and medical imaging, anatomically
realistic and biophysically detailed multi-scale computer
models of the heart are playing an increasingly impor-
tant role in advancing our understanding of integrated
cardiac function in health and disease [4]. For example,
action potential propagation simulation of ischemia in 3D
anatomically detailed ventricle is valuable in studying the
mechanisms and dynamics of ischemia-induced re-entry
and arrhythmia [5].
Cardiac electrophysiological model generally is a cou-
pled system of partial differential equations for modeling
the electrical wave propagation across tissue and ordinary
differential equations for modeling cell dynamics. The
solution to typical electrophysiological model, e.g., bido-
main and monodomain model, usually involves millions
of nodes, and suffers from the complex boundary and
computational inefficiency problems. Moreover, the multi-
CPU-based high-performance computing is also ineffi-
cient for 3D data visualization. To address the complex
boundary problem, a phase-field method was proposed in
[6], and Lu et al. adopted it for studying the influence
of ischemia on 3D human ventricle [7]. To alleviate the
computational inefficiency, high-performance computing
was developed to speedup electrophysiological simula-
tion. However, the scalability of most bidomain based
simulation is limited to hundreds of cores, and only few
work circumvented this limitation by using appropriate
grid partitioning and asynchronous IO parallel mechanism
[8], [9].
With the advent of GPU, general-purpose computing
on GPUs (GPGPU) becomes an emerging technology
to enhance computational efficiency [10]–[12]. Instead
of CPUs, GPU has massively parallel single instruction
multiple data processing units with hundreds of stream
processors, and these architectural advantages can be
utilized to speedup cardiac electrical activity simulation
[13], [14]. Most recently, Nimmagadda et al. proposed
a bidomain model with multi-GPU implementation for
electrophysiological simulation on clinical time-scales
[15].
JOURNAL OF COMPUTERS, VOL. 9, NO. 2, FEBRUARY 2014
doi:10.4304/jcp.9.2.360-367