Probabilistic Anatomical Connectivity Using Completion Fields

Parya MomayyezSiahkal, Kaleem SiddiqiMICCAI 2010 · DOI

Abstract

Quantifying anatomical connectivity from Diffusion MRI (dMRI) remains a significant challenge due to noise and the heuristic nature of traditional tractography. This paper presents a framework based on 3D Stochastic Completion Fields, modeling connectivity as a probability distribution governed by the Fokker-Planck equation. By employing an unconditionally stable numerical scheme and integrating local fiber orientation data, the method provides a robust connectivity index that overcomes the distance-dependent decay typical of Monte-Carlo methods. Validation on the MICCAI 2009 Fiber Cup phantom demonstrates superior consistency and specificity compared to standard probabilistic tractography.

Introduction

In neuroimaging, Diffusion MRI allows researchers to observe the movement of water molecules within the brain. Because water diffuses more readily along axons than across them, this data is used to map anatomical connectivity through a process known as tractography. However, existing methods struggle to quantify the strength of these connections.

Deterministic methods often ignore noise, while probabilistic methods—which typically rely on Monte-Carlo simulations—frequently depend on heuristics like voxel visit counts. These counts tend to drop off sharply with distance, making long-range connections appear artificially weak. The objective of this work is to establish a mathematically principled connectivity index that is robust to noise and capable of handling the complex, twisting geometries of neural fiber paths.

Methods

The proposed framework utilizes 3D Stochastic Completion Fields to model the probability of a particle traveling between a source and a sink region.

The Fokker-Planck Equation

The core of the model is the Fokker-Planck equation, a partial differential equation (PDE) that describes the evolution of a probability distribution over time, accounting for both advection (forward motion) and diffusion (random spreading).

Numerical Stability via Operator Splitting

Solving these equations in the context of complex brain geometry is numerically difficult. To ensure stability, the authors implement an operator-splitting scheme:

  • Lax-Wendroff scheme: Used for advection components.
  • Crank-Nicholson scheme: Used for diffusion components. This approach remains stable even for highly curved tracts, such as cortical U-fibers, without requiring arbitrary curvature thresholds.

Local Data Integration

Rather than using a uniform model, the algorithm tunes diffusion parameters ($\sigma$) and decay rates ($\zeta$) at each voxel based on the Fiber Orientation Distribution (FOD) derived from the MRI data. This ensures that the probability "flow" is guided by the underlying biological structure.

Experimental Results

The algorithm was validated using the MICCAI 2009 Fiber Cup phantom, a physical mock-up of brain fibers made of hydrophobic acrylic. This phantom provides a known "ground truth" to evaluate the algorithm's ability to identify true connections while ignoring false ones.

Figure 1: The MICCAI 2009 Fiber Cup phantom and the six test scenarios (a-f) used to evaluate connectivity.

The researchers compared the Stochastic Completion Field (SCF) against a state-of-the-art Probabilistic Tractography baseline across six scenarios (labeled a-f in Figure 1).

Key Findings

  • Consistency: For true connections, the SCF provided connectivity indices in a stable range (0.17 to 0.24). In contrast, the baseline results fluctuated wildly, dropping as low as 0.00008.
  • Robustness: In scenario (d)—a straight fiber tract—the baseline algorithm failed entirely (returning 0), while the SCF correctly identified the connection with high confidence (0.24).
  • Specificity: In scenarios (e) and (f), which represent regions not physically connected, the SCF correctly assigned near-zero values, demonstrating high specificity.

Discussion and Future Work

While the SCF method shows significant improvements in robustness and distance-handling, it faces a "Bottleneck Problem." In cases where a true fiber tract forms a bottleneck with another pathway (as seen in scenario c), the algorithm may occasionally favor a smoother, incorrect path (scenario f) because the underlying PDE prefers smoothness over sharp directional changes.

Future research will focus on utilizing higher-resolution MRI data to better distinguish fibers in crowded regions and incorporating "curvature consistency" constraints to help the algorithm maintain the correct trajectory through crossing or merging fiber bundles.

Summary updated

BibTeX
@inproceedings{momayyezsiahkal2010probabilistic,
  author = {Parya MomayyezSiahkal and Kaleem Siddiqi},
  title = {{Probabilistic Anatomical Connectivity Using Completion Fields}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2010},
  doi = {10.1007/978-3-642-15705-9\_69},
}
Figure