Statistical shape modeling of multi-organ anatomies with shared boundaries

Krithika Iyer, Alan Morris, Brian Zenger, Karthik Karanth, Nawazish Khan, Benjamin Orkild, Oleksandre Korshak, Shireen ElhabianFrontiers in Bioengineering and Biotechnology 2023 · DOI

TL;DR

This paper introduces a novel workflow for statistical shape modeling (SSM) that explicitly accounts for shared boundaries between interconnected organs, such as the chambers of the heart. By extracting and modeling contact surfaces as independent entities, the researchers eliminated common artifacts like overlapping structures or artificial gaps, allowing for more accurate identification of pathological changes in complex anatomical systems.

What Problem Does This Paper Address?

Statistical shape modeling (SSM) is a computational method used to quantify anatomical variation across a population. Traditionally, SSM treats organs as isolated objects. However, many critical anatomical structures are "multi-organ" systems with shared boundaries—think of the interventricular septum between the heart's ventricles or the cartilage between joints.

When standard SSM methods are applied to these systems, they often fail to maintain anatomical integrity. Because the interaction between surfaces isn't explicitly modeled, the resulting statistical "modes of variation" might show organs bleeding into one another (particle overlap) or pulling apart (unrealistic gaps). These artifacts make it difficult for clinicians to distinguish between a modeling error and a genuine pathological change, such as the flattening of the heart septum due to pressure overload.

Methods

The authors build upon Particle-Based Shape Modeling (PSM), a state-of-the-art approach that uses a system of "correspondence particles" to describe shapes. Their contribution is a specialized pipeline to handle shared boundaries:

  1. Shared Boundary Extraction: The team developed a tool to detect where two meshes are in close proximity. They extract this contact area as a separate shared surface ($M$) and its outline as a shared contour ($C$).
  2. Preprocessing: They use isotropic re-meshing and Laplacian smoothing to ensure the extracted surfaces are high-quality and free of jagged artifacts.
  3. Modified Objective Function: The core "math" of PSM involves minimizing an entropy-based energy function to distribute particles across surfaces. The authors modified this function to include the shared surface and contour. Crucially, particles on the shared contour act as a "buffer," repelling particles from the adjacent organs to prevent them from crossing into the wrong territory.
  4. Multi-level Component Analysis (MLCA): To ensure that differences in how organs are positioned (pose) didn't get confused with how they are shaped, the authors used MLCA to disentangle these two types of variation.

Datasets

The method was validated on two distinct datasets:

  • Synthetic "Peanut" Dataset: A controlled set of 30 shapes (15 controls, 15 "pathological") consisting of two spheres where one is subtracted from the other. This provided a "ground truth" to test if the model could correctly identify known shape changes without creating overlaps.
  • Cardiac Biventricular Dataset: A clinical dataset featuring MRIs from 6 healthy volunteers and 23 patients with various cardiovascular conditions (e.g., congestive heart failure, pulmonary hypertension). The focus was on the right ventricle (RV), left ventricle wall (LVW), and the interventricular septum (IVS).

Key Insights and Ideas

  • The "Buffer Zone" Concept: By treating the shared contour as a repelling force, the model naturally maintains a "clean" boundary. This prevents the statistical model from suggesting that one organ can physically occupy the same space as another.
  • Clinical Relevance of Curvature: The model's ability to maintain a consistent parameterization of the interventricular septum allowed it to capture septal curvature as a primary mode of variation. In cardiology, a "flattened" septum is a known indicator of right ventricular pressure overload.
  • Disentangling Pose from Shape: The use of MLCA was vital. In clinical data, organs aren't always perfectly aligned during imaging. Without MLCA, the model might mistakenly flag a slight rotation of the heart as a structural deformity.

Key Results

  • Elimination of Artifacts: Compared to "multiple-domain" models (which model organs separately but simultaneously), the proposed shared-boundary model successfully eliminated the gaps and overlaps visible in the first five dominant modes of variation.
  • Pathology Identification: In the cardiac dataset, the model clearly identified a decrease in interventricular septal curvature in the patient group compared to healthy controls. This aligns perfectly with established clinical observations of structural remodeling in heart disease.
  • Consistency: The shared boundary model achieved a more consistent distribution of particles across the population, leading to a more "compact" and statistically valid representation of the anatomy.

Limitations and Future Work

The authors acknowledge a few areas for improvement:

  • Manual Tuning: The threshold for detecting what constitutes a "shared boundary" currently requires experimental tuning for different datasets.
  • Sample-Level Extraction: The boundary extraction currently happens on a per-sample basis. The authors suggest that a population-level approach—extracting boundaries based on the statistics of the entire cohort—might be more robust.
  • Gaussian Assumptions: The current model assumes a Gaussian distribution for shape variations. Future iterations could incorporate non-linear modeling to capture more complex, non-Gaussian anatomical changes.

By explicitly modeling the "handshake" between organs, this research moves SSM closer to a truly automated, artifact-free tool for clinical diagnostic support.

Summary updated

Abstract

<b>Introduction:</b> Statistical shape modeling (SSM) is a valuable and powerful tool to generate a detailed representation of complex anatomy that enables quantitative analysis of shapes and their variations. SSM applies mathematics, statistics, and computing to parse the shape into some quantitative representation (such as correspondence points or landmarks) which can be used to study the covariance patterns of the shapes and answer various questions about the anatomical variations across the population. Complex anatomical structures have many diverse parts with varying interactions or intricate architecture. For example, the heart is a four-chambered organ with several shared boundaries between chambers. Subtle shape changes within the shared boundaries of the heart can indicate potential pathologic changes such as right ventricular overload. Early detection and robust quantification could provide insight into ideal treatment techniques and intervention timing. However, existing SSM methods do not explicitly handle shared boundaries which aid in a better understanding of the anatomy of interest. If shared boundaries are not explicitly modeled, it restricts the capability of the shape model to identify the pathological shape changes occurring at the shared boundary. Hence, this paper presents a general and flexible data-driven approach for building statistical shape models of multi-organ anatomies with shared boundaries that explicitly model contact surfaces. <b>Methods:</b> This work focuses on particle-based shape modeling (PSM), a state-of-art SSM approach for building shape models by optimizing the position of correspondence particles. The proposed PSM strategy for handling shared boundaries entails (a) detecting and extracting the shared boundary surface and contour (outline of the surface mesh/isoline) of the meshes of the two organs, (b) followed by a formulation for a correspondence-based optimization algorithm to build a multi-organ anatomy statistical shape model that captures morphological and alignment changes of individual organs and their shared boundary surfaces throughout the population. <b>Results:</b> We demonstrate the shared boundary pipeline using a toy dataset of parameterized shapes and a clinical dataset of the biventricular heart models. The shared boundary model for the cardiac biventricular data achieves consistent parameterization of the shared surface (interventricular septum) and identifies the curvature of the interventricular septum as pathological shape differences.
BibTeX
@inproceedings{iyer2023statistical,
  author = {Krithika Iyer and Alan Morris and Brian Zenger and Karthik Karanth and Nawazish Khan and Benjamin Orkild and Oleksandre Korshak and Shireen Elhabian},
  title = {{Statistical shape modeling of multi-organ anatomies with shared boundaries}},
  booktitle = {Frontiers in Bioengineering and Biotechnology},
  year = {2023},
  doi = {10.3389/fbioe.2022.1078800},
}
Figure