Abstract
Computational models of cardiac electromechanics (EM) are increasingly being applied to clinical problems, with patient-specific models being generated from high fidelity imaging and used to simulate patient physiology, pathophysiology and response to treatment. Current structured meshes are limited in their ability to fully represent the detailed anatomical data available from clinical images and capture complex and varied anatomy with limited geometric accuracy. In this paper, we review the state of the art in image-based personalization of cardiac anatomy for biophysically detailed, strongly coupled EM modeling, and present our own tools for the automatic building of anatomically and structurally accurate patient-specific models. Our method relies on using high resolution unstructured meshes for discretizing both physics, electrophysiology and mechanics, in combination with efficient, strongly scalable solvers necessary to deal with the computational load imposed by the large number of degrees of freedom of these meshes. These tools permit automated anatomical model generation and strongly coupled EM simulations at an unprecedented level of anatomical and biophysical detail.
Introduction
The human heart is a complex multi-physics system where electrical signals (electrophysiology) trigger mechanical contractions, which subsequently influence electrical activity through mechano-electric feedback. Accurately simulating this "strongly coupled" behavior is critical for clinical applications, such as predicting patient responses to cardiac resynchronization therapy. Historically, researchers have relied on "weak coupling," solving electrical waves on fine grids while calculating mechanical movement on coarser, simplified structured meshes. However, these structured meshes often fail to capture fine-scale anatomical features like endocardial trabeculae and papillary muscles, limiting their predictive power. This paper presents an automated pipeline for creating high-fidelity, patient-specific cardiac models using high-resolution unstructured meshes to overcome these geometric limitations.
Methodology
The proposed approach utilizes high-resolution unstructured tetrahedral meshes to solve both electrophysiology and mechanics simultaneously. The automated pipeline consists of several integrated steps designed to move from raw medical imaging to a functional biophysical model.
Key components of the pipeline include:
- Automated Segmentation: Delineating heart chambers and tissues from MRI or CT data.
- Variational Smoothing: Removing "staircase" artifacts from segmented images while preserving total anatomical volume.
- Mesh Generation: Utilizing the Tarantula tool to generate boundary-fitted tetrahedral meshes capable of resolving features down to approximately 110 µm.
- Rule-Based Fiber Assignment: Implementing the Laplace-Dirichlet method to mathematically assign muscle fiber orientations based on histological data, as individual fibers are not visible in standard clinical imaging.
- High-Performance Computing (HPC) Solvers: Leveraging strongly scalable solvers and algebraic multigrid preconditioners to manage the massive computational load associated with millions of degrees of freedom.
Datasets and Experimental Setup
The pipeline's versatility was demonstrated using two distinct data sources:
- Ex vivo Rabbit Ventricles: High-resolution MRI data (~25 µm isotropic resolution) was used to showcase the capture of intricate endocardial structures.
- In vivo Human Heart: A 3D steady-state free precession (SSFP) MRI of a whole human heart at 1.3 mm resolution was used to construct a clinical-scale four-chamber model.
Discussion: The Resolution Paradox and Mesh Selection
A key insight from this work is the "resolution paradox": while mechanics were traditionally thought to require less resolution than electrophysiology, high resolution is actually necessary to accurately capture complex anatomy, such as the thin walls of the right ventricle. By adopting unstructured meshes over template-based structured meshes, the models can adapt to unique, often pathological, patient-specific geometries without labor-intensive manual adjustments. Furthermore, the study emphasizes that strong scaling is mandatory; solving problems with nearly 100 million degrees of freedom requires software capable of scaling across thousands of CPU cores.
Key Results
The primary achievement of this research is the simulation of a single heartbeat in a high-resolution four-chamber human heart model. The mesh comprised 184.6 million tetrahedral elements, resulting in 95.9 million displacement degrees of freedom. Running on 8,192 compute cores of the SuperMUC supercomputer, a single heartbeat simulation was completed in approximately 235.3 minutes. The resulting model successfully captured the heterogeneity of fiber stretch and calcium distribution, illustrating how anatomical architecture directly influences electrical re-entry risks.
Conclusion and Future Directions
While this work represents a significant technical advancement in cardiac modeling, several challenges remain for clinical adoption. The computational resources and time (nearly 4 hours on a supercomputer) are currently prohibitive for routine hospital use. Future efforts will focus on streamlining the pipeline, utilizing GPU acceleration, and improving model validation. Additionally, better methods for data assimilation—integrating real-time clinical measurements like ECGs or pressure-volume loops—are needed to further personalize and tune these models for individual patients.
Summary updated