Fisher-Rao Regularized Transport Analysis of the Glymphatic System and Waste Drainage

Rena Elkin, Saad Nadeem, Hedok Lee, Helene Benveniste, Allen R. TannenbaumMICCAI 2020 · DOI

TL;DR

This paper introduces a novel Lagrangian framework for analyzing the complex, time-varying dynamics of the glymphatic system, the brain's waste removal pathway. By employing Fisher-Rao regularization within an optimal mass transport (OMT) model, they achieve a unified visualization of particle trajectories and attributes, enabling clearer differentiation of flow patterns under different physiological conditions, such as varying anesthetic states.

What Problem Does This Paper Address?

The glymphatic system (GS) plays a crucial role in clearing waste products from the brain, and its dysfunction is linked to neurodegenerative diseases like Alzheimer's. Understanding and quantifying GS transport is vital for brain health. While optimal mass transport (OMT) has been used for image analysis, its dynamic application for complex biological flows, especially those with subtle diffusion, has been challenging. Existing Eulerian methods struggle to capture the full temporal dynamics of individual particles, and previous OMT regularization techniques can be sensitive to noise. This paper aims to provide a more robust and interpretable method for analyzing these dynamic flows, specifically to differentiate GS function under different states of arousal and anesthesia.

Methods

The core of this work is the application of a Fisher-Rao regularized optimal mass transport (OMT) framework. While the standard OMT problem seeks to minimize the cost of transforming one probability distribution into another, this paper focuses on the dynamical OMT formulation. They adopt a specific regularization approach that adds a Fisher-Rao information term to the kinetic energy of the system. This is presented as an equivalent, and in some ways more interpretable, alternative to the Eulerian-based Schrödinger bridge OMT regularization.

The key innovation lies in using a Lagrangian framework to visualize the results. Instead of looking at flow properties at fixed points in space (Eulerian), this approach tracks individual particles over time. This allows for a unified visualization of particle trajectories (pathlines) and their associated attributes (like density and speed) in a single image. This unified representation is particularly useful for distinguishing subtle differences in flow patterns that might be missed with Eulerian methods.

Datasets

The study utilizes data from Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI) acquired from rats. Specifically, they analyze data from rats under two different anesthetic conditions:

  1. DEXM+ISO: A combination of dexmedetomidine and low-dose isoflurane, mimicking "light" sleep/hypnosis.
  2. ISO: Deep anesthesia induced by isoflurane.

The data resolution is noted as 128x128x128. The authors mention that the initial and final densities for the OMT problem are taken from these DCE-MRI datasets.

Key Insights and Ideas

  • Unified Lagrangian Visualization: The Fisher-Rao regularized OMT framework, when coupled with a Lagrangian visualization approach, provides a powerful way to represent complex, time-varying biological flows. Seeing particle trajectories and attributes together offers a more intuitive understanding of fluid dynamics.
  • Differentiating Anesthetic States: The method successfully distinguishes between GS flow patterns under different anesthetic conditions. This highlights how subtle physiological changes can significantly alter brain waste clearance.
  • Glymphatic System Insights: The analysis reveals that GS transport is more efficient under DEXM+ISO anesthesia (mimicking natural sleep) compared to deep ISO anesthesia. This suggests that "natural" sleep states may be superior for maintaining GS function.
  • Olfactory Efflux as a "Valve": Paradoxically, in ISO anesthesia, solute efflux into the nasal conchae appears more vigorous, potentially acting as an "efflux valve" when overall fluid flow is impaired. This points to compensatory mechanisms in waste clearance.

Key Results

  • The Lagrangian analysis successfully captures known GS flow patterns and differentiates between the two anesthetic states.
  • More pathlines are observed in the DEXM+ISO anesthetized rats, indicating accelerated GS transport and greater solute movement into the parenchyma and along the middle cerebral artery (MCA).
  • In ISO anesthesia, while parenchymal uptake is reduced, there's a paradoxical increase in pathline speed in the nasal conchae, suggesting a different clearance route is being utilized.
  • Quantitative analysis (Figure 4E) confirms a higher proportion of faster-moving particles in the clustered pathlines in ISO anesthetized rats, particularly towards the nasal conchae.

Limitations and Future Work

  • Small ε Values: The paper notes that for small values of the diffusion parameter (ε), standard Sinkhorn algorithms become unstable, necessitating the use of the Fisher-Rao approach. This implies that the method is specifically designed for scenarios where diffusion is minimal but important.
  • Future Work: The authors plan to use the computed particle attributes (like speed and density) as features for machine learning algorithms. This will enable them to differentiate and classify various transport-related flow patterns for diagnostic purposes.

Summary updated

BibTeX
@inproceedings{elkin2020fisher,
  author = {Rena Elkin and Saad Nadeem and Hedok Lee and Helene Benveniste and Allen R. Tannenbaum},
  title = {{Fisher-Rao Regularized Transport Analysis of the Glymphatic System and Waste Drainage}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2020},
  doi = {10.1007/978-3-030-59728-3\_56},
}
Figure