Editor’s note: This article is by Dr. Jeremy Rice a
computational physiologist at IBM’s Thomas J. Watson Research Center, as told
to Chris Nay, IBM Research Communications.
The Journal of the American College of Cardiology reported
in 2006 that about two of 1,000 people, worldwide, die of a ventricular
arrhythmia every year – the most common cause of sudden cardiac death. Predicting
who will die suddenly from a ventricular arrhythmia is a huge challenge, but
current computer simulations that could help cardiologists find effective
therapies take hours or even days to run a single heartbeat.
To address this need for faster simulations, my team at
IBM Research is working with the Lawrence Livermore National Lab on Cardioid, a
code that simulates the human heart on the 20 Petaflop Blue Gene/Q, Sequoia. The
96-rack installation at LLNL can run Cardioid roughly 1,200 times faster than
other published results to simulate in exquisite detail the electrophysiology
of up to three billion heart cells (similar to the number in real heart) and their
cell-to-cell electric coupling.
Our hope is that cardiac specialists could eventually
model how their patients will react to certain drugs and how genetic variations
predispose some patients to arrhythmias.
What’s happening in the heart
Cardioid runs fast enough to allow scientific inquires
that were previously impractical with the existing modeling platforms. For
example, we have the potential to model natural variations seen across a real
population of patients.
Every heart beat electrically excites every heart cell
(unlike skeletal muscle that recruits more or fewer muscles cells depending on
the demands). Each heart cell is a small excitable system, and electrical
activation spreads in a salutatory fashion to its neighbors.
But during a heart beat, there isn’t a clear separation
between the activation of a cell and the interaction between the cells. Hence, the
problem of a heart simulation needs a two-fold solution that requires a
computer with the ability to track the individual cells and their interactions,
up to 10,000 times for each heart beat (the number of interactions between
cells, per second). This inter-processor communication is exactly what Sequoia
So, what does Cardiod, running on Sequoia, reveal?
The heart beats at a constantly changing rate, and we
found that arrhythmia occurs much more at elevated drug concentrations and at
slower heart rates (termed bradycardia), or abnormally skipped heart beats. Our
mathematical models can complement clinical work to allow cardiac specialists
and researchers to extract more data from experimental studies.
A heart outside the body
“In 1946, Dr. John Gibbon, who built the first
heart-lung machine 9 years earlier and went on to perform the first human
heart bypass operation in 1953, began working with then-president of IBM Thomas
Watson Sr. and five IBM engineers to build an improved heart-lung machine.”
Experiments don’t let us look inside a beating heart to
see what is happening at the level of the individual cells. But our simulations
let us predict the individual cellular responses. In fact, our simulations
suggest arrhythmias can arise from subtle and complex cellular interactions
that cannot be resolved with experimental techniques, now or in the foreseeable
Building a billion heart cells
Models of single cardiac cells
have been developed since the 1960s, and three-dimensional anatomic (or
geometric) models of various cardiac structures have been developed since the
1990s. And our work, with the University
of Rochester, last year showed
that models can greatly complement experimental and clinical data to reveal
more about how genetic variations affect a person’s susceptibility to
arrhythmia. However, this work only modeled single cells or small networks of cells
that lack the complexity of the whole heart.
IBM & LLNL
IBM and LLNL
announced the Deep Computing Solutions collaboration. It gives businesses
access to HPC systems and to IBM and LLNL scientists and engineers for
simulations like Cardiod’s.
Our work to model the entire heart actually started in
1998, harnessing Sequoia’s early predecessor, the Blue Gene/L. Matthias
Reumann, a post-doctoral student at IBM’s Thomas J Watson Research Lab at the
time (now at IBM Research – Australia), developed a code to decompose the heart
into small pieces and distribute the work to up to 32,000 processors – an
unprecedented number at the time.
Cardioid is today’s version of that code. It can scale to
more than 100 times the number of cores than the original – and runs about
13,000 times faster overall on Sequoia. This power means simulations of between
180 million and three billion heart cells, depending on the level of spatial
detail we want to model.
Cardioid runs fast enough to allow scientific inquires that were previously
impractical with the existing modeling platforms. We have the potential to model
natural variations seen across a real population of patients. For example, we
could measure drug effects highly dependent on individual characteristics such
as age, gender, and disease history.
Also, patients’ drug concentrations constantly change
from the time the drug is administered until it eventually leaves their bodies
(or another dose is administered). Simulating these time scales were simply
impracticable before Cardioid.
We hope to see Cardioid used by companies in the
pharmaceutical industry and medical device companies to help with clinical
decision support in treating deadly diseases, such as arrhythmias and
congestive heart failure. Longer-term applications could include virtual drug
trials over simulated patient population.
This video shows how a medication can impact the heart to
promote arrhythmias. Here we see a block of the heart wall in a common
experimental preparation, with the simulated drug on the left and control case
on the right. The red color shows activated tissue, and the blue shows
recovered tissue. The interface between red and blue shows a wavefront of
activation that spreads from cell to cell. The two sequential stimulations
(applied at the location of the boxes on the left surface) produce an
arrhythmia known as a spiral wave, where the wavefront can continually spin.
Without drug, the pattern will die out quickly whereas with the drug, the cycle
is sustained and is potentially fatal.