Unbalanced Regularized Optimal Mass Transport with Applications to Fluid Flows in the Brain

Xinan Chen, Helene Benveniste, Allen R. TannenbaumarXiv 2023 · DOI · arXiv

Introduction

This research presents a significant advancement in computational fluid dynamics for medical imaging, specifically targeting the "leaky" nature of biological systems. Classical physics models often rely on strict mass-conservation requirements, which are frequently violated in medical imaging contexts like Dynamic Contrast-Enhanced MRI (DCE-MRI). In these scenarios, tracers are constantly entering and exiting the field of view, causing signal intensities to fluctuate. To address this, the authors introduce unbalanced regularized Optimal Mass Transport (urOMT), a mathematical framework capable of tracking fluid flow even when the total mass changes over time.

Methodology: The urOMT Framework

Standard Optimal Mass Transport (OMT) models seek the most efficient way to move a distribution of mass from one state to another under the constraint that mass must be conserved. The urOMT model relaxes this constraint by incorporating several key technical components:

  • The Source Term ($r$): A "relative source" variable is added to the standard advection-diffusion equation, acting as a local sink or source where mass can appear or disappear.
  • Fisher-Rao Metric: To prevent the model from simply "teleporting" mass to minimize costs, a Fisher-Rao term penalizes the creation or destruction of mass. This creates a mathematical balance between transport and source activity.
  • Free End-Point Condition: To account for noise in real-world MRI data, a "free-urOMT" version is implemented. This includes a fitting term ($\beta$) that allows for slight discrepancies between the model and the final image, preventing over-fitting.
The discrete-time framework of urOMT showing transitions between density states $\rho_i$ with source and transport terms across time steps $t_0$ to $t_m$.

The parameter $\alpha$ acts as a critical "dial" in this system. A high $\alpha$ forces the model toward standard mass-conserving OMT, while a low $\alpha$ allows for more significant mass fluctuations, enabling researchers to tune the model based on the known "leakiness" of the biological system being studied.

Experimental Validation and Results

The model was validated using both synthetic data and 3D MRI scans of a healthy rat brain. The synthetic "toy" dataset consisted of Gaussian spheres that moved diagonally while simultaneously diffusing and changing in brightness, serving as a proof-of-concept for the model's ability to handle mass appearing and disappearing.

Validation on synthetic Gaussian spheres showing input images (a), time-averaged Eulerian relative-source maps (b, d), speed maps (c, e), and various flow visualizations including pathlines (f) and Péclet-lines (i).

In the rat brain application, the model provided the first precise quantification of tracer movement in the glymphatic system. Key insights included:

  • Glymphatic Mapping: The model visualized tracer entry into the cerebrospinal fluid (CSF), its movement into perivascular spaces (the brain's "highways"), and its eventual penetration into deep brain tissue.
  • Advection vs. Diffusion: By calculating the Péclet number along flow paths, the researchers proved that transport in large vessels is advection-dominated (driven by flow), whereas transport in deep tissue is dominated by diffusion.
  • Eulerian and Lagrangian Perspectives: The framework provides both "speed maps" (Eulerian) and "pathlines" (Lagrangian), offering a dual view essential for understanding both local clearance rates and global flow patterns.

Conclusion and Future Work

urOMT successfully captures the complex fluid dynamics of the brain, including the symmetrical entry of fluid from the midline of the skull base. While the current model assumes a constant diffusion coefficient, future iterations may incorporate spatially dependent diffusion terms to better reflect the heterogeneous geometry of the brain's extracellular matrix. This framework represents a vital step toward quantifying the brain's mysterious waste-clearance pathways and has broad implications for studying neurodegenerative diseases.

Summary updated

Abstract

As a generalization of the optimal mass transport (OMT) approach of Benamou and Brenier's, the regularized optimal mass transport (rOMT) formulates a transport problem from an initial mass configuration to another with the optimality defined by the total kinetic energy, but subject to an advection-diffusion constraint equation. Both rOMT and the Benamou and Brenier's formulation require the total initial and final masses to be equal; mass is preserved during the entire transport process. However, for many applications, e.g., in dynamic image tracking, this constraint is rarely if ever satisfied. Therefore, we propose to employ an unbalanced version of rOMT to remove this constraint together with a detailed numerical solution procedure and applications to analyzing fluid flows in the brain.
BibTeX
@misc{chen2023unbalanced,
  author = {Xinan Chen and Helene Benveniste and Allen R. Tannenbaum},
  title = {{Unbalanced Regularized Optimal Mass Transport with Applications to Fluid Flows in the Brain}},
  year = {2023},
  eprint = {2301.11228},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2301.11228},
}
Figure