World’s second most powerful supercomputer goes straight for the heart

Cardioid simulation shows how a drug might affect heart function. Side-by-side images compare a portion of a beating heart with (left column) and without (right column) the administered drug. An electric stimulus applied at 0 milliseconds causes heart cells to depolarize (red). Before repolarization is complete (blue), an unplanned stimulus causes a premature depolarization (at 560 milliseconds). In the simulation with the drug, a reentrant circulation pattern develops (see circle at 690 milliseconds), inhibiting repolarization of the cells and thus preventing the heart from beating normally.

One of the most powerful supercomputers on Earth, the Sequoia at Lawrence Livermore National Laboratory (LLNL), California, US, has simulated the fastest and most detailed model of a beating human heart yet. This will aid scientists researching the side effects of pacemakers and drugs that may cause deadly arrhythmia -- abnormal electrical activity linked to 325,000 deaths every year in the US. This research was done swiftly, before the supercomputer is classified for work by the US government on its nuclear weapons stockpile in 2013.   

Sequoia is an IBM BlueGene/Q system, which made it into first place this June in the Top500 supercomputer list with a performance of 16.32 petaflops (quadrillions of floating operating points) per second, on the LINPACK benchmark using 1,572,864 computer cores. When fully operational, one hour of Sequoia’s computer power is expected to be equal to 6.7 billion people working with hand calculators on a calculation 24 hours per day, 365 days a year, for 320 years.

A heart simulation at the cellular level

Researchers at Lawrence Livermore National Laboratory and the IBM Research, US, used a software code called Cardioid to simulate electrophysiological activity – or electrical connections -- within and among cells in a human heart at a resolution of an unprecedented scale - a spatial resolution of a heart cell, which is about 0.1 millimeters long.

The simulated heart was divided up into a large number of manageable pieces, or subdomains, with processing of 3,800 heart cells assigned to a single computing node. A total of 370 million cells were simulated. The Cardioid code was scalable, meaning that its performance increased in proportion to the number of cores applied to the problem.

A three-dimensional level of detail was achieved by combining cross-sectional images from the Visible Human Project, from the National Library of Medicine, Maryland, US, which contains detailed 3D anatomical datasets of men and women. Another piece of software was developed to reconstruct the anatomy of a human torso, which means Cardioid offers multiscale simulation, from sub-cellular functions up to clinical signals from actual patients.

Simulating one hour of normal heart activity -- thousands of heart beats -- took the Sequoia system seven hours. Previous codes took 45 minutes just to simulate a single heartbeat. Now the model could be a possible test bed for clinical trials of drugs and pacemakers that affect heart rates, before being tested on humans. Physicians could study drug affects in a simulated heart over hours, which was not possible before.

Heart models for real-world applications

In a recent paper, published in the journal Science Translational Medicine,written by four professors affiliated with Johns Hopkins Institute for Computational Medicine, in Maryland, US, say that computational models of electrical activity in the heart are on their way to being used to guide doctors in preventing sudden cardiac death and diagnosing and treating those at risk.

“The implementation of the heart model on Sequoia is a major breakthrough in all the application areas of computational medicine discussed in our paper, and has implications for all model-based methods for personalized care at the point of treatment,” says Raimond Winslow, director of the Johns Hopkins Institute for Computational Medicine, who was not involved in the Cardioid modeling.

This includes surgical planning in cancer tumor resections, optimal placement of deep brain stimulation electrodes, and shape analysis of brains and hearts for early detection and diagnosis of disease. “I'm also proud to say that one of the team members at IBM T.J. Watson, Jeremy Rice, did his PhD research in my laboratory,” says Winslow.

The Cardioid code divides the heart into a large number of manageable pieces, or subdomains. The development team used two approaches, called Voronoi (left) and grid (right), to break the enormous computing challenge into much smaller individual tasks.

Winslow says Sequoia’s technology will impact the speed of simulations and if it can be made more widespread, then simulations lasting minutes can be used in an interactive fashion at the point of care or provide offline comparisons along with 3D shapes of the heart, for targeted ablation treatments on the heart. 

Now, studies are being carried out on the integration of Cardiod’s electrophysiological simulations with mechanical components that model pumping of blood and the contractions of the heart, branching out the code’s ability to address therapies for diseases such as congestive heart failure.

One ambitious idea mentioned in LLNL’s Science and Technology Review publication is to merge the Cardioid simulation with a patient’s clinical data, such as electrocardiograms, magnetic resonance imaging, and computed tomography scans. Leader of the Lawrence Livermore team, Art Mirin says, "Our planned comprehensive whole heart modeling capability, which includes vascular and chamber blood flow as well as electrophysiology and mechanics, will enable customization to better reflect individual patients' needs, which should lead to superior devices (for example, cardiac pacemakers) and pharmaceuticals (such as anti-arrhythmic drugs), and hence save lives."

This goal of a multi-scale computer model for personalized medicine is mirrored in Europe by the Virtual Physiological Human Network of Excellence. Peter Kohl, a biophysicist from Imperial College London and co-founder of Virtual Physiological Human was quoted saying an accurate model of the heart requires 10²⁶ possible space-time combinations or 100 million, billion, billion data points, which only large-scale distributed computing or supercomputers can handle.

Researchers involved in the Cardioid simulation achievement were selected as finalists of the Association of Computing Machinery Gordon Bell prize award, that recognizes the highest accomplishments in supercomputing, unveiled this week at the Supercomputing Conference 2012 in Salt Lake City, Utah, US.