Template‐based landmark and region mapping of bone

Jaeil Kim, Sang Gyo Seo, Dong Yeon Lee, Jinah ParkJournal of Foot and Ankle Research 2014 · DOI

TL;DR

This paper introduces an automated method for mapping anatomical landmarks and regions onto 3D bone models using non-rigid template registration. This approach significantly speeds up and improves the consistency of bone shape analysis, which is crucial for understanding biomechanics and diagnosing conditions like deformities.

What Problem Does This Paper Address?

Analyzing bone shape using 3D surface models is becoming increasingly important in biomechanics research. It allows for quantitative assessments of bone deformities and their correlation with joint motion. To perform these analyses, researchers need to identify specific anatomical landmarks and regions on the bone surface. However, manually identifying these features on individual 3D bone models is a tedious, time-consuming, and error-prone process due to variations in bone size and shape across subjects. This paper aims to automate this process, making bone shape analysis more efficient and reliable.

Methods

The core of the proposed method is a non-rigid template-to-image registration technique. A generic 3D triangular mesh model of the target bone (in this case, the calcaneus) is created, which also encodes predefined anatomical landmarks and regions as a subset of its mesh points. This template model is then non-rigidly deformed to fit individual 3D bone scans. The deformation is guided by a Laplacian deformation framework. This framework ensures that the transformation minimizes distortion of the point distribution within the template, effectively "tracing" the positions of the anatomical landmarks and regions from the generic template onto the specific individual bone model.

Datasets

The authors used CT scans for their experiments. They constructed a calcaneus template model from manually segmented CT scans. The method was then applied to segmentations of three subjects with different calcanei shapes. The paper does not mention the creation or release of a new benchmark dataset.

Key Insights and Ideas

  • Automating a Bottleneck: The manual identification of anatomical landmarks on 3D bone models is a significant bottleneck in shape analysis. Automating this process unlocks the potential for larger-scale studies and more consistent analyses.
  • Laplacian Deformation for Robustness: The use of a Laplacian deformation framework is key. It allows for flexible fitting to individual bone shapes while preserving the underlying structure and relationships between landmarks, preventing unrealistic distortions.
  • Template as a "Guide": The generic template acts as a knowledgeable guide, encoding anatomical knowledge that is then transferred to individual scans, even when those scans have significant shape variations.

Key Results

The method demonstrated high accuracy in reconstructing individual bone shapes, achieving a mean volume overlap of 0.981±0.009 and a mean distance of 0.349±0.423 mm compared to manual segmentations. Crucially, the automatically identified landmarks showed strong consistency with manually assigned landmarks by three blinded operators, suggesting the method reliably captures anatomical features. The figure visually confirms that the automatic landmark and region mapping successfully aligns with the bone structures of different subjects.

Limitations and Future Work

The authors acknowledge that the current study was performed on a limited number of subjects. They plan to assess the robustness and accuracy of their automatic landmark and region mapping method using larger datasets. This suggests that while promising, further validation is needed to confirm its generalizability across a wider range of bone morphologies and potential pathologies.

Summary updated

Abstract

Background The shape morphology using 3D surface models has been recently emerged for biomechanics research, such as the quantitative assessment of bone deformity with clinical factors [1] and the correlation analysis between bone shape and joint motion [2]. In the bone shape morphology, the morphological difference of the bones across subjects is quantified by the geometric measures, such as the curvature of the articular surface and the relative bone orientation in joints, defined with the anatomical landmarks and regions on the bone surface. However, the landmark and region determination on individual cases is a difficult and time-consuming task, because of the various size and shape of the bones and operator’s errors. In this paper, we propose an automated landmark and region mapping method based on a non-rigid template-toimage registration. The template model is a triangular mesh including the generic shape of the target. It also encodes the landmarks and regions as a subset of the points in the triangular mesh. For the landmark and region mapping to individual bones, the template model is nonrigidly deformed by a Laplacian deformation framework [3]. This framework derives the point transformation into the image boundary while minimizing the distortion of the point distribution in the template model. This behavior of the deformation framework helps to trace the positions of the anatomical landmarks and regions across subjects.
BibTeX
@article{kim2014templatebased,
  author = {Jaeil Kim and Sang Gyo Seo and Dong Yeon Lee and Jinah Park},
  title = {{Template‐based landmark and region mapping of bone}},
  journal = {Journal of Foot and Ankle Research},
  year = {2014},
  doi = {10.1186/1757-1146-7-s1-a42},
}
Figure