Barycentric Subspace Analysis: A New Symmetric Group-Wise Paradigm for Cardiac Motion Tracking

Marc-Michel Rohé, Maxime Sermesant, Xavier PennecMICCAI 2016 · DOI

TL;DR

This paper introduces "Barycentric Subspace Analysis" (BSA) for cardiac motion tracking, moving beyond single-image references to a more robust subspace representation. This novel approach improves tracking accuracy, particularly for large deformations like those between end-diastole and end-systole, and enhances the estimation of crucial clinical metrics like ejection fraction.

What Problem Does This Paper Address?

Accurately tracking cardiac motion is vital for diagnosing and monitoring heart conditions. Traditional methods often rely on registering each frame of a cardiac sequence to a single reference image, typically the end-diastole (ED) frame. This can introduce biases, especially when quantifying large deformations that occur during the cardiac cycle, such as the transition from end-systole (ES) to ED. The challenge is to develop a registration framework that is less sensitive to the choice of a single reference and can better capture the complex, dynamic nature of cardiac motion.

Methods

The core innovation is the Barycentric Subspace Analysis (BSA). Instead of a single reference image, BSA defines a subspace spanned by multiple reference images. This subspace, a "Barycentric Template," is parameterized by barycentric coordinates, which represent the projection of a given image onto this subspace.

Key methodological contributions include:

  • Barycentric Subspaces: Generalizing the concept of principal component analysis to manifolds, these subspaces are defined by a set of reference images and their corresponding barycentric coefficients.
  • Barycentric Template: A novel representation of cardiac motion that is not a single image but a set of barycentric coordinates, allowing for a more flexible and group-wise analysis.
  • Group-wise Analysis: The paper demonstrates that these barycentric coefficients can serve as a meaningful "signature" for cardiac motion, effectively distinguishing between healthy subjects and patients with cardiac conditions.
  • Relaxed Regularization for Registration: BSA is used as a prior to relax the regularization term in cardiac motion tracking. Instead of regularizing the entire deformation field, regularization is applied only to the component representing the image's distance to the barycentric subspace. This allows for better capture of large deformations.
  • Symmetric and Transitive Registration: The framework is extended to develop a symmetric and transitive registration method, improving robustness and reducing error accumulation over the sequence.

Datasets

  • STACOM 2011 cardiac motion tracking challenge dataset: Used for evaluating the cardiac motion signature for group-wise analysis. This dataset consists of 15 healthy subjects.
  • Tetralogy of Fallot (ToF) patients dataset: Used for group-wise analysis, comprising 10 patients.
  • Synthetic time series of cardiac image frames: Used for evaluating the registration improvement. This synthetic dataset provides ground truth meshes for accurate comparison.

Key Insights and Ideas

  • Subspaces as References are More Robust: Using a subspace spanned by multiple reference images is inherently more robust than relying on a single reference, mitigating biases associated with the choice of that single image.
  • Barycentric Coordinates as a "Cardiac Signature": The barycentric coordinates derived from BSA provide a powerful and discriminative representation of cardiac motion dynamics, capable of separating different patient populations.
  • Relaxing Regularization for Large Deformations: By applying regularization only to the "residual" part of the deformation (distance to the subspace), the method can better accommodate and track large deformations, which are critical for accurate cardiac function assessment.
  • Improved Clinical Metric Estimation: The enhanced tracking accuracy directly translates to more precise estimation of clinically important parameters like ejection fraction.

Key Results

  • Improved Ejection Fraction (EF) Estimation: The barycentric registration method reduced the estimation error of the ejection fraction by approximately 50% compared to standard methods, bringing it closer to the ground truth.
  • Reduced Tracking Error for Large Deformations: The barycentric registration achieved around 30% smaller error for large deformations (e.g., between ED and ES) compared to standard methods.
  • Effective Group Separation: The barycentric coefficients showed significant differences between healthy subjects and ToF patients, demonstrating their utility as a cardiac motion signature for group-wise analysis.
  • Symmetric Registration Improvement: The symmetric-barycentric extension reduced the maximum error over the sequence by approximately 10% compared to standard barycentric registration.

Limitations and Future Work

  • Computational Cost: While not explicitly detailed as a limitation, the introduction of subspaces and more complex calculations might imply increased computational demands compared to simpler methods.
  • Choice of Reference Images: The paper discusses finding "optimal references" but the process and its sensitivity could be a point for further investigation.
  • Future Work: The authors suggest that the methodology can be extended for symmetric transitive registration, which is beneficial for more robust tracking along the entire cardiac sequence. Further exploration of the clinical applicability and validation on larger, more diverse patient cohorts would be valuable.

Summary updated

BibTeX
@inproceedings{rohe2016barycentric,
  author = {Marc-Michel Rohé and Maxime Sermesant and Xavier Pennec},
  title = {{Barycentric Subspace Analysis: A New Symmetric Group-Wise Paradigm for Cardiac Motion Tracking}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2016},
  doi = {10.1007/978-3-319-46726-9\_35},
}
Figure