Large scale cardiac modeling on the Blue Gene supercomputer.

Area of Science:

• Computational biology
• Biophysics
• Medical imaging

Background:

• Current multi-scale, multi-physical heart models face computational limitations, hindering accuracy and resolution.
• Achieving high spatial resolution in cardiac models is computationally intensive.
• Existing models struggle to balance model detail with computational feasibility.

Purpose of the Study:

• To propose a novel framework for computing large-scale cardiac models with enhanced accuracy and resolution.
• To address computational limitations in simulating detailed cardiac electrophysiology.
• To enable faster and more efficient simulations of cardiac function.

Main Methods:

• Utilized optimal recursive bisection (ORB) for anatomical data decomposition and parallel distribution.
• Implemented monodomain equations for the diffusion term, incorporating heterogeneous anisotropy.
• Employed the Visible Female dataset for ventricles at 0.2 mm resolution on an IBM Blue Gene/L parallel computer.

Main Results:

• Reduced simulation time from 87 minutes (512 nodes) to 11 minutes (8192 nodes) for a 10 ms simulation.
• Demonstrated that load balancing was effective across computational nodes despite data distribution variations.
• Achieved near-optimal speedup for computation time, with communication overhead being the primary limitation.

Conclusions:

• The developed framework enables the simulation of detailed cardiac models within hours, facilitating a single heartbeat simulation.
• This approach overcomes previous computational barriers, paving the way for more complex and accurate cardiac modeling.
• The study highlights the potential of parallel computing for advancing cardiac electrophysiology research.