Smoothed Particle Hydrodynamics for Electrophysiological Modeling: An Alternative to Finite Element Methods

Èric Lluch, Rubén Doste, Sophie Giffard-Roisin, Alexandre This, Maxime Sermesant, Oscar Camara, Mathieu De Craene, Hernán G. MoralesFIMH 2017 · DOI

TL;DR

This paper introduces Smoothed Particle Hydrodynamics (SPH) as a novel, meshless alternative to traditional Finite Element Methods (FEM) for simulating cardiac electrophysiology. The authors demonstrate that SPH can accurately model electrical propagation in the heart, even in the presence of complex geometries and pathological conditions like scars, while potentially simplifying the simulation pipeline by eliminating the need for mesh generation.

What Problem Does This Paper Address?

Simulating the electrical activity of the heart (electrophysiology) is crucial for understanding its function and diagnosing/treating diseases. Traditional methods, like Finite Element Methods (FEM), are widely used but face a significant bottleneck: the generation of complex, high-quality computational meshes required to represent the heart's intricate geometry. This meshing process is time-consuming, non-trivial, and can hinder the rapid translation of these models into clinical practice. Furthermore, coupling electrophysiology with mechanics in FEM can introduce complexities due to differing mesh requirements. This paper seeks to overcome these limitations by exploring a meshless approach.

Methods

The core of this work is the application of Smoothed Particle Hydrodynamics (SPH) to solve the Mitchell-Schaeffer (MS) electrophysiological model. SPH is a meshless Lagrangian method where the simulation domain is represented by a set of discrete particles, each carrying its own properties. Instead of relying on a mesh, SPH approximates continuous functions and their derivatives by summing contributions from neighboring particles, weighted by a kernel function.

The authors adapted the MS model, which describes the action potential duration (APD) and includes a diffusion term accounting for fiber orientation, for SPH discretization. They specifically used a cubic B-spline kernel for its favorable properties in approximating derivatives. To address potential issues with boundary conditions in SPH, they also implemented a corrective smoothed particles method (CSPM) to enhance accuracy near boundaries. The SPH approach was compared against a standard FEM implementation of the same model.

Datasets

Not discussed in the paper. The study uses an image-based left ventricular geometry extracted from medical imaging.

Key Insights and Ideas

  • Meshless Advantage: The most significant insight is that SPH offers a viable alternative to FEM by completely bypassing the need for mesh generation. This could dramatically streamline the process of creating patient-specific cardiac models from imaging data.
  • Accuracy Comparable to FEM: The SPH approach demonstrated comparable accuracy to FEM in simulating electrical depolarization times, with differences generally within a few milliseconds. This suggests SPH can capture the essential physics of cardiac electrical propagation.
  • Pathological Modeling Potential: The paper successfully demonstrates SPH's ability to model a pathological scenario, specifically a scar in the heart wall. This highlights its potential for simulating arrhythmias and other scar-related conditions.
  • Computational Trade-offs: While SPH eliminates meshing, the computational cost can be higher than FEM, especially with larger kernel sizes or higher particle resolutions. However, the authors suggest that GPU implementation could mitigate this.

Key Results

  • Depolarization Time Comparison: SPH and FEM simulations showed similar depolarization times, with maximum differences of up to 10.9% at the apex.
  • Qualitative Agreement: Visual comparisons of activation patterns and depolarization maps between SPH and FEM showed good qualitative agreement.
  • Pathological Simulation: In a scar scenario, SPH simulations accurately depicted altered electrical activation patterns, including delayed activation in low-conductivity regions and potential reentry arrhythmias in zero-conductivity scar zones.
  • Computational Time: FEM was generally faster than the SPH implementation on a CPU. For example, with 18667 particles and a 0.005m kernel size, SPH took 3 minutes 13 seconds, while FEM took 9.92 seconds. However, the SPH approach's computational cost increased significantly with larger kernel sizes.

Limitations and Future Work

  • Full Heart Geometries: The authors acknowledge that their study was limited to a left ventricular geometry and plan to extend their work to full heart models to evaluate SPH's robustness.
  • Validation with Patient Data: Further validation of SPH for electrophysiology is needed by comparing simulation results with actual patient data, not just validated synthetic geometries.
  • Impact of Particle Distribution: The influence of particle distribution on the results needs to be further investigated, particularly when particle motion is considered.
  • Computational Efficiency: While SPH is promising, improving its computational efficiency, potentially through GPU acceleration, is a key area for future work.

Summary updated

BibTeX
@inproceedings{lluch2017smoothed,
  author = {Èric Lluch and Rubén Doste and Sophie Giffard-Roisin and Alexandre This and Maxime Sermesant and Oscar Camara and Mathieu De Craene and Hernán G. Morales},
  title = {{Smoothed Particle Hydrodynamics for Electrophysiological Modeling: An Alternative to Finite Element Methods}},
  booktitle = {FIMH},
  year = {2017},
  doi = {10.1007/978-3-319-59448-4\_32},
}
Figure