Group-Wise Consistent Fiber Clustering Based on Multimodal Connectional and Functional Profiles

Bao Ge, Lei Guo, Tuo Zhang, Dajiang Zhu, Kaiming Li, Xintao Hu, Junwei Han, Tianming Liu

Abstract

Diffusion Tensor Imaging (DTI) tractography generates vast amounts of white matter fiber data, making individual and group-wise analysis challenging. This paper presents a novel two-stage hybrid approach to organize these fibers into consistent, biologically meaningful bundles. By integrating structural connectivity from DTI with functional signals from resting-state fMRI (rsfMRI), the method identifies 32 group-wise consistent fiber bundles. The approach transitions from identifying a reliable structural "backbone" to a functional classification stage, significantly increasing fiber coverage while maintaining high functional coherence and geometric similarity across subjects.

Introduction

Mapping the brain's white matter "wiring" via DTI tractography is a cornerstone of modern neuroscience. However, the sheer volume of data—often exceeding 100,000 fibers per subject—necessitates effective fiber clustering techniques. Traditional methods primarily rely on geometric similarity, often failing to ensure consistency across different individuals or ignoring the functional relationships between the brain regions being connected. This study addresses these gaps by proposing a multimodal framework that leverages both structural landmarks and functional synchronization.

Methodology

The proposed pipeline operates in two distinct stages to move from high-confidence structural anchors to a broader classification of the fiber population.

The proposed two-stage multimodal fiber clustering pipeline, integrating DTI and rsfMRI data.

Stage 1: Establishing the Structural Backbone

The process begins with 358 validated cortical landmarks, known as Dense Individualized and Common Connectivity-based Cortical Landmarks (DICCCOLs). Fibers connecting these landmarks across 37 Brodmann areas are identified to form 32 "backbone" bundles. These represent the most anatomically consistent fibers present across the cohort.

Stage 2: Functional Classification

To classify the remaining fibers, the authors utilize rsfMRI data. The Maximal Overlap Discrete Wavelet Transform (MODWT) is applied to extract functional signals at fiber endpoints, focusing on the 0.04–0.08 Hz frequency band where functional brain activity is most prominent. Fibers are then assigned to a backbone bundle if their functional coherence (correlation of signals at both ends) aligns with the established group profile.

Results

The methodology was validated using a dataset of 8 subjects scanned on a 3T MRI system. The two-stage approach successfully identified 32 consistent bundles in every subject.

Visualization of group-wise consistency across 8 subjects, showing (a) all clustered fibers and (b) representative fibers for each bundle.

Key quantitative findings include:

  • Increased Coverage: The initial backbone accounted for only 8.03% of total fibers. The functional classification stage expanded this to 25.78% of the total fiber population.
  • Functional Coherence: The resulting bundles exhibited high functional correlation values, ranging from 0.73 to 0.90, suggesting strong synchronization between connected regions.
  • Geometric Consistency: Low Hausdorff distances between representative fibers across subjects confirmed high physical similarity, validating the "group-wise consistent" nature of the clustering.

Discussion and Future Work

By shifting the paradigm from unsupervised clustering to a guided classification based on a structural backbone, this method provides a robust way to handle the complexity of tractography data. The integration of rsfMRI ensures that the identified bundles are not just geometrically similar but also functionally relevant.

Limitations include the current landmark coverage, which leaves approximately 74% of fibers unclassified because the landmarks do not yet cover the entire brain surface. Future work will focus on expanding the landmark set and validating these structural bundles against task-based fMRI to confirm their involvement in specific cognitive and motor functions.

Summary updated

Abstract

Fiber clustering is an essential step towards brain connectivity modeling and tract-based analysis of white matter integrity via diffusion tensor imaging (DTI) in many clinical neuroscience applications. A variety of methods have been developed to cluster fibers based on various types of features such as geometry, anatomy, connection, or function. However, identification of group-wise consistent fiber bundles that are harmonious across multi-modalities is rarely explored yet. This paper proposes a novel hybrid two-stage approach that incorporates connectional and functional features, and identifies group-wise consistent fiber bundles across subjects. In the first stage, based on our recently developed 358 dense and consistent cortical landmarks, we identified consistent backbone bundles with representative fibers. In the second stage, other remaining fibers are then classified into the existing backbone bundles using their correlations of resting state fMRI signals at the two ends of fibers. Our experimental results show that the proposed methods can achieve group-wise consistent fiber bundles with similar shapes and anatomic profiles, as well as strong functional coherences.
BibTeX
@inproceedings{ge2012group,
  author = {Bao Ge and Lei Guo and Tuo Zhang and Dajiang Zhu and Kaiming Li and Xintao Hu and Junwei Han and Tianming Liu},
  title = {{Group-Wise Consistent Fiber Clustering Based on Multimodal Connectional and Functional Profiles}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2012},
  doi = {10.1007/978-3-642-33454-2\_60},
}
Figure