An automatic service for the personalization of ventricular cardiac meshes

Pablo Lamata, Matthew Sinclair, Eric Kerfoot, Angela Lee, Andrew Crozier, Bojan Blazevic, Sander Land, Adam J. Lewandowski, David C. Barber, Steven Niederer, Nic SmithJournal of The Royal Society Interface 2013 · DOI

TL;DR

This paper introduces "heartgen," a fully automated cloud-based service that generates personalized, high-quality computational meshes of the heart's ventricles from segmented medical images. This addresses a major bottleneck in computational cardiac physiology, enabling more accurate and accessible patient-specific simulations for disease management and research.

What Problem Does This Paper Address?

Computational modeling of the heart holds immense promise for understanding and treating cardiovascular diseases. However, a significant hurdle has been the laborious and complex process of tailoring these computational models to individual patients' unique anatomy and physiology. Specifically, creating accurate and stable "computational meshes" – the geometric frameworks upon which simulations are run – from patient scans has been a time-consuming manual task. Existing methods often struggle with high-order interpolation (needed for accurate mechanical simulations), robustness across different patient data, and minimizing the impact of segmentation errors. This paper aims to bridge this gap by providing an automated, versatile, and robust solution for generating these crucial cardiac meshes.

Methods

The "heartgen" service employs a four-step pipeline:

  1. Shape Analysis for a Tailored Template Mesh: Instead of using a single generic template, the system first analyzes the input binary mask of the patient's anatomy to create a customized idealized template. This involves identifying key anatomical features like the left and right ventricular blood pools and estimating the heart's spatial orientation and dimensions. This tailored template significantly improves the subsequent registration accuracy.
  2. Image Registration and Level of Detail (LoD): The service uses the Sheffield Image Registration Toolkit (ShIRT) to align the patient's anatomy with the tailored template. A key innovation here is the introduction of a "Level of Detail" (LoD) parameter, allowing users to balance fitting accuracy with simulation stability. Lower LoD values are more robust for sparse data or segmentation errors, while higher LoD values prioritize anatomical accuracy.
  3. Mesh Warping: Once the deformation field is established through registration, a variational technique is used to warp the high-order interpolation mesh (cubic Hermite or Lagrange) to conform to the patient's anatomy. This method deforms the entire continuum, not just individual nodes, leading to smoother and more stable meshes.
  4. Quality Enhancement: A post-processing step linearizes the transmural degrees of freedom within the mesh elements. This improves mesh quality and regularity without sacrificing anatomical accuracy, which is crucial for the stability of mechanical simulations.

The entire service is deployed on a cloud infrastructure, making it accessible via a web service.

Datasets

The paper evaluates the "heartgen" service using:

  • Two statistical atlases: Used for segmentation and to generate average human biventricular (BiV) anatomy models (A1, A2).
  • A cohort of 216 left ventricular (LV) and 39 BiV cases: These were manually segmented from short-axis MRI images. LV cases were derived from dynamic MRI studies, while BiV cases used whole-heart MRI images.
  • Examples of varying anatomical information: Including dilated/normal hearts, different species, and datasets ranging from sparse (6/8 slices) to high-resolution animal data.

The paper does not introduce a new benchmark dataset but rather demonstrates the service's performance on existing types of cardiac imaging data.

Key Insights and Ideas

  • Tailored Templates are Key: The realization that a single generic template is insufficient and that customizing the template based on the input anatomy dramatically improves registration convergence and mesh quality is a significant insight.
  • The LoD Trade-off: Explicitly introducing and quantifying the Level of Detail (LoD) parameter as a user-controllable trade-off between anatomical accuracy and simulation stability is a practical and important contribution for users.
  • High-Order Meshes for Mechanics: The paper strongly advocates for and demonstrates the utility of high-order interpolation meshes (cubic Hermite/Lagrange) for accurate cardiac mechanics simulations, highlighting their efficiency and improved accuracy over linear schemes.
  • Robustness to Imperfect Data: The method's ability to minimize the impact of segmentation errors and handle sparse data (e.g., from dynamic studies) makes it highly valuable for real-world clinical applications where data quality can vary.
  • Cloud Deployment for Accessibility: Making this sophisticated meshing tool available as a cloud-based web service democratizes access to advanced computational modeling capabilities for the scientific community.

Key Results

  • High Success Rate: The service achieved fully automatic personalization with accuracy and quality requirements met in 87% of 255 cases.
  • Versatility: Successfully applied to cardiac geometries with different pathologies, various species, and diverse data resolutions (from sparse to high-resolution).
  • Accuracy and Quality: For the tested cohorts, median fitting errors were 1.18 mm (LV) and 1.62 mm (BiV), with median mesh qualities (Jacobian ratio) of 0.37 for both.
  • Performance with Sparse Data: Demonstrated gradual degradation in accuracy and quality with increasing inter-slice distance, but still provided usable results.
  • Quality Enhancement: The post-processing step improved mesh quality by an average of 8.7% without significant loss in accuracy.
  • Computational Cost: Personalization takes between 5 and 25 minutes per case on a conventional desktop machine.

Limitations and Future Work

  • Anatomical Correspondence: The proposed method does not guarantee exact anatomical correspondence between different cases. Future work could involve incorporating fiducial anatomical markers (e.g., RV-LV insertion points, valve planes) to improve this.
  • Topology Limitations: The structured BiV topology has limitations at the RV apex, where the anatomical shape is not perfectly captured.
  • Generalization of Robustness Thresholds: The thresholds for accuracy and quality used in the robustness analysis are based on empirical studies and may need re-evaluation for different applications.
  • Extension to Other Structures: The authors suggest that the service can be generalized to include other anatomical structures (atria, entire heart) or different topological requirements, which would involve defining suitable templates and alignment algorithms.
  • Broader Applications: While optimized for mechanical simulations, the authors note that mesh requirements for electrophysiology or blood flow simulations differ, suggesting future work to adapt the methodology for these areas.

Summary updated

Abstract

Computational cardiac physiology has great potential to improve the management of cardiovascular diseases. One of the main bottlenecks in this field is the customization of the computational model to the anatomical and physiological status of the patient. We present a fully automatic service for the geometrical personalization of cardiac ventricular meshes with high-order interpolation from segmented images. The method is versatile (able to work with different species and disease conditions) and robust (fully automatic results fulfilling accuracy and quality requirements in 87% of 255 cases). Results also illustrate the capability to minimize the impact of segmentation errors, to overcome the sparse resolution of dynamic studies and to remove the sometimes unnecessary anatomical detail of papillary and trabecular structures. The smooth meshes produced can be used to simulate cardiac function, and in particular mechanics, or can be used as diagnostic descriptors of anatomical shape by cardiologists. This fully automatic service is deployed in a cloud infrastructure, and has been made available and accessible to the scientific community.
BibTeX
@article{lamata2013automatic,
  author = {Pablo Lamata and Matthew Sinclair and Eric Kerfoot and Angela Lee and Andrew Crozier and Bojan Blazevic and Sander Land and Adam J. Lewandowski and David C. Barber and Steven Niederer and Nic Smith},
  title = {{An automatic service for the personalization of ventricular cardiac meshes}},
  journal = {Journal of The Royal Society Interface},
  year = {2013},
  doi = {10.1098/rsif.2013.1023},
}
Figure