Fiber direction estimation, smoothing and tracking in diffusion MRI

Raymond Chi-Wing Wong, Thomas C. M. Lee, Debashis Paul, Jie Peng, Alzheimer’s Disease Neuroimaging InitiativeThe Annals of Applied Statistics 2016 · DOI · arXiv

Abstract

Diffusion magnetic resonance imaging is an imaging technology designed to probe anatomical architectures of biological samples in an in vivo and noninvasive manner through measuring water diffusion. The contribution of this paper is threefold. First, it proposes a new method to identify and estimate multiple diffusion directions within a voxel through a new and identifiable parametrization of the widely used multi-tensor model. Unlike many existing methods, this method focuses on the estimation of diffusion directions rather than the diffusion tensors. Second, this paper proposes a novel direction smoothing method which greatly improves direction estimation in regions with crossing fibers. This smoothing method is shown to have excellent theoretical and empirical properties. Last, this paper develops a fiber tracking algorithm that can handle multiple directions within a voxel. The overall methodology is illustrated with simulated data and a data set collected for the study of Alzheimer's disease by the Alzheimer's Disease Neuroimaging Initiative (ADNI).

Introduction

Diffusion magnetic resonance imaging (dMRI) is a powerful noninvasive tool for mapping the white matter tracts of the brain by tracking the movement of water molecules. The traditional gold standard, Diffusion Tensor Imaging (DTI), operates on the assumption of a single primary diffusion direction per voxel. However, neuroanatomy is complex; in many regions, distinct fiber bundles cross within the same voxel. In these areas, DTI often fails, producing isotropic "blobs" or incorrect directions that lead to fragmented or biased connectivity maps. While multi-tensor models have been developed to address this, they frequently suffer from "non-identifiability," where multiple mathematical configurations can explain the same signal, leading to unstable results. Furthermore, many advanced methods require high-resolution data (HARDI) that is often unavailable in clinical settings.

The DiST Framework

This paper introduces DiST (Diffusion Direction Smoothing and Tracking), a comprehensive framework designed to resolve complex fiber architectures using standard clinical MRI data. The methodology consists of three integrated steps:

1. Identifiable Voxel-wise Estimation

The authors re-parametrize the multi-tensor model to focus specifically on the principal diffusion directions (eigenvectors) rather than the full diffusion tensors. This mathematical shift ensures the model is identifiable even with a fixed b-value (a standard MRI setting), allowing for the stable estimation of up to three directions per voxel.

2. Spatial Smoothing via Clustering

A major challenge in dMRI is that standard spatial smoothing often blurs crossing fibers into a single, incorrect average direction. DiST overcomes this with a "cluster-then-smooth" approach. It first clusters the directions of neighboring voxels into homogeneous groups using the PAM algorithm and silhouette scores. It then applies a weighted Karcher mean—a specialized method for averaging points on a sphere—only within those clusters. This preserves the distinct "lanes" of crossing traffic while effectively reducing noise.

Visualization of estimated fiber directions showing complex crossing patterns and the resolution of multiple pathways within voxels.

3. Advanced Fiber Tracking

The framework evolves the popular FACT (Fiber Assignment by Continuous Tracking) algorithm. The DiST version is capable of handling multiple directions per voxel and incorporates "skip" logic: if the algorithm encounters a voxel with no clear direction (due to noise or isotropy), it can look one voxel ahead to determine if the path continues. This significantly reduces the occurrence of short, fragmented tracts.

Experimental Validation

The DiST methodology was validated using both simulated data and a real-world dataset from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). The ADNI scan utilized 41 gradient directions on a 3 Tesla scanner, a setup typical of modern clinical research rather than specialized high-end experimental physics, demonstrating the method's practical utility.

Results and Discussion

A critical test for DiST involved the intersection of the corpus callosum (CC) and the corona radiata (CR). Standard DTI typically shows the CC being "blocked" or terminated by the CR. In contrast, DiST successfully reconstructed both pathways passing through each other, matching established neuroanatomical atlases. In the ADNI sample, DiST identified two distinct fiber directions in over 50% of the voxels in the studied region (593 out of 1125 voxels), highlighting the prevalence of crossing fibers that single-tensor models fail to capture. Simulation results further confirmed that the smoothing procedure significantly reduces error in direction estimation, particularly at acute crossing angles.

Conclusion

DiST offers a robust and clinically viable solution for mapping complex brain connectivity. By prioritizing direction estimation and employing a novel spatial clustering technique, it overcomes the stability and resolution limitations of traditional dMRI methods. Future research will focus on extending the tracking to a probabilistic framework and applying DiST to large patient cohorts to investigate correlations between improved fiber maps and cognitive decline in neurodegenerative diseases.

Summary updated

Abstract

Diffusion magnetic resonance imaging is an imaging technology designed to probe anatomical architectures of biological samples in an in vivo and noninvasive manner through measuring water diffusion. The contribution of this paper is threefold. First, it proposes a new method to identify and estimate multiple diffusion directions within a voxel through a new and identifiable parametrization of the widely used multi-tensor model. Unlike many existing methods, this method focuses on the estimation of diffusion directions rather than the diffusion tensors. Second, this paper proposes a novel direction smoothing method which greatly improves direction estimation in regions with crossing fibers. This smoothing method is shown to have excellent theoretical and empirical properties. Last, this paper develops a fiber tracking algorithm that can handle multiple directions within a voxel. The overall methodology is illustrated with simulated data and a data set collected for the study of Alzheimer's disease by the Alzheimer's Disease Neuroimaging Initiative (ADNI).
BibTeX
@article{wong2016fiber,
  author = {Raymond Chi-Wing Wong and Thomas C. M. Lee and Debashis Paul and Jie Peng and Alzheimer’s Disease Neuroimaging Initiative},
  title = {{Fiber direction estimation, smoothing and tracking in diffusion MRI}},
  journal = {The Annals of Applied Statistics},
  year = {2016},
  doi = {10.1214/15-aoas880},
}
Figure