Introduction
In Diffusion MRI (dMRI), mapping the complex architecture of neuronal fibers is essential for understanding brain connectivity. However, standard techniques frequently struggle with intravoxel fiber crossings—regions where multiple fiber bundles intersect within a single 3D volume element (voxel). While measures like Generalized Fractional Anisotropy (GFA) can indicate overall order, they often fail to distinguish between a single coherent fiber and multiple crossing populations. This paper introduces Fibre Population Dispersion (FPD), a robust statistical metric designed to quantify fiber partitioning without prior assumptions about the number of fibers present.
Methods
The proposed framework utilizes the Spherical Polar Fourier (SPF) expansion to define FPD as the normalized autocorrelation of a spherical feature function. By rotating the diffusion signal and measuring its self-similarity, the method identifies structural complexity within the voxel.
To ensure computational efficiency, the authors derived a closed-form solution using the Wiener-Khinchin theorem and Wigner-D functions. This allows FPD to be calculated directly from signal coefficients via a fast dot product, bypassing slow numerical simulations. Furthermore, the use of unit quaternions provides a proper distance metric on the rotation group, avoiding mathematical artifacts like gimbal lock. The resulting 3D dispersion is summarized into a scalar index by calculating the $p$-th statistical moment, typically the second moment ($p=2$), to balance contrast and noise resistance.
Validation and Datasets
The FPD metric was validated across three distinct datasets:
- Synthetic Data: Simulated signals for two fiber populations crossing at specific angles (0.5 and 1.37 radians).
- MICCAI 2009 Fibre Cup Phantom: A physical phantom made of hydrophobic acrylic filaments, providing a "ground truth" for crossing angles.
- Rat Spinal Cord Phantom: A biological model created by crossing two excised rat spinal cords in agar, testing the algorithm on real tissue architecture.
Key Results and Insights
The core innovation of FPD is using autocorrelation as a proxy for underlying structure. A voxel containing crossing fibers produces a fundamentally different autocorrelation pattern than one with a single fiber population.
Experimental results demonstrated several key advantages:
- Superior Crossing Detection: In the Fibre Cup phantom, FPD clearly highlighted crossing regions that appeared indistinguishable from single-fiber regions in standard GFA maps.
- Sensitivity to Geometry: The FPD index was found to be proportional to the crossing angle, effectively ranking the complexity of the underlying tissue.
- Biological Accuracy: In the rat spinal cord test, FPD successfully identified the specific 3D crossing point where one cord curved over another, a region where standard anisotropy measures failed to show differentiation.
Conclusion and Future Work
FPD provides a granular, local statistical analysis of intravoxel fiber architecture. By operating within the spherical harmonics domain, it transforms a complex integration problem into an efficient algebraic one. Future research will focus on translating this tool to clinical human data to detect early signs of neurological diseases, such as Multiple Sclerosis, where fiber dispersion patterns may change. Additionally, integrating FPD maps into tractography algorithms could significantly improve navigation through complex white matter crossings.
Summary updated