Modeling of Personalized Anatomy Using Plastic Strains

Bohan Wang, George Matcuk, Jernej Barbic

TL;DR

This paper introduces a novel method for creating personalized 3D anatomical models from medical scans by deforming a generic template using "plastic strains." This approach effectively handles the large, spatially varying, and anisotropic deformations often seen in human anatomy, overcoming limitations of previous methods that produced artifacts like spikiness and wiggliness.

What Problem Does This Paper Address?

Reconstructing accurate, personalized 3D anatomical models from medical imaging data (like MRI or CT scans) is crucial for applications ranging from surgical planning and medical education to realistic human simulation in computer graphics and virtual reality. A key challenge is that individual anatomy varies significantly, requiring models to undergo large, non-uniform, and anisotropic shape changes. Existing methods often struggle with these complex deformations, producing undesirable artifacts. This paper aims to bridge this gap by developing a robust shape deformation technique capable of capturing these intricate anatomical variations.

Methods

The core of the paper is a new shape deformation method that leverages plastic strains within the Finite Element Method (FEM) framework. Instead of relying solely on elastic forces, the method models large deformations by computing "plastic deformation gradients" for each element (tetrahedron) in the mesh. This allows for arbitrary, spatially varying stretching and compression without the inherent stability issues of purely elastic models.

The optimization process involves finding these plastic strains and the resulting vertex positions of the mesh to match sparse point constraints (landmarks and ICP markers) derived from medical images. Key technical contributions include:

  • Plastic Strain Formulation: Using plastic deformation gradients as the primary mechanism for large, anisotropic shape changes.
  • Gauss-Newton Optimization: Adapting a Gauss-Newton method to efficiently solve the highly nonlinear optimization problem, overcoming numerical challenges.
  • Handling Attachments and Free-Flying Objects: Developing techniques to manage objects that are attached to other structures (like muscles to bones) and those that are un-attached (like a liver), including a novel method for solving linear systems with singular matrices.
  • Robustness to Sparse Constraints: Demonstrating that the method performs well even with a limited number of input markers, which is common in medical imaging.

Datasets

The paper uses medical imaging data (MRI and CT scans) of various human anatomical structures for evaluation. Specific examples include:

  • Human Hand Muscles: Extracted from MRI scans.
  • Liver: Extracted from CT scans.
  • Hip Bone: Extracted from CT scans.
  • Hip Muscle (Gluteus Medius): Extracted from MRI scans.

The authors also reference publicly available datasets for template models, such as those from the University of Dundee, Zygote, and Ziva Dynamics.

Key Insights and Ideas

  • Plasticity as the Key to Large Deformations: The paper convincingly argues that plastic strains are a more natural and effective tool than purely elastic models for capturing the large, anisotropic, and spatially varying deformations characteristic of human anatomy. This is a significant departure from traditional graphics approaches.
  • Sparse Constraints are Sufficient: The method's ability to achieve high accuracy with sparse point constraints (landmarks and ICP markers) is a major advantage, as obtaining dense, accurate correspondences from medical images is often difficult and time-consuming.
  • Unified Framework for Attached and Un-attached Objects: The development of a single framework that can handle both attached and free-flying objects simplifies the modeling process and broadens the applicability of the technique.
  • Avoiding Artifacts: The plastic strain approach inherently avoids the spikiness and wiggliness that plague traditional elastic deformation methods when dealing with large strains, leading to more visually plausible and anatomically accurate results.

Key Results

  • High Accuracy: The method consistently matches medical image constraints to within 0.5 mm or better, which is at or beyond the accuracy of the MRI scanners used.
  • Successful Modeling of Diverse Anatomy: The approach was successfully applied to model various anatomical structures, including complex hand muscles, a liver, a hip bone, and a hip muscle, demonstrating its versatility.
  • Superiority over Existing Methods: Comparisons to standard methods like ARAP, BBW, LBW, FEM, and variational methods show that the plastic strain approach produces significantly smoother and more accurate results, especially in challenging scenarios with large, anisotropic strains.
  • Efficiency: While not the primary focus, the paper presents efficient numerical techniques (e.g., adapted Gauss-Newton, solving singular systems) that enable the method to handle complex models. For instance, solving a single linear system for the hand muscle example took 9.5 seconds with their method, compared to 897.5 seconds with a naive conjugate gradient approach.

Limitations and Future Work

  • Tet Mesh Quality: The authors acknowledge that if plastic strains cause tetrahedra to collapse, numerical instabilities can arise. While not a problem in their examples, they suggest that re-meshing during optimization could address this.
  • Template Mesh Requirement: The method requires a plausible, non-degenerate template tet mesh. Future work could involve developing methods to start optimization from simpler shapes like spheres, potentially through re-meshing.
  • Scalability with Markers: The method's performance will slow down with a very large number of markers, as the size of the dense matrix to invert increases.
  • Extension to Shells: The authors suggest that the plastic strain deformation framework could be extended to model shells (like cloth) by formulating the appropriate elastic energy and derivatives.

Summary updated

Abstract

We give a method for modeling solid objects undergoing large spatially varying and/or anisotropic strains, and use it to reconstruct human anatomy from medical images. Our novel shape deformation method uses plastic strains and the Finite Element Method to successfully model shapes undergoing large and/or anisotropic strains, specified by sparse point constraints on the boundary of the object. We extensively compare our method to standard second-order shape deformation methods, variational methods and surface-based methods and demonstrate that our method avoids the spikiness, wiggliness and other artefacts of previous methods. We demonstrate how to perform such shape deformation both for attached and un-attached ("free flying") objects, using a novel method to solve linear systems with singular matrices with a known nullspace. While our method is applicable to general large-strain shape deformation modeling, we use it to create personalized 3D triangle and volumetric meshes of human organs, based on MRI or CT scans. Given a medically accurate anatomy template of a generic individual, we optimize the geometry of the organ to match the MRI or CT scan of a specific individual. Our examples include human hand muscles, a liver, a hip bone, and a gluteus medius muscle ("hip abductor").
BibTeX
@article{wang2021modeling,
  author = {Bohan Wang and George Matcuk and Jernej Barbic},
  title = {{Modeling of Personalized Anatomy Using Plastic Strains}},
  journal = {ACM Trans. Graphics},
  year = {2021},
  doi = {10.1145/3443703},
}
Figure