A Novel Deep Learning Tractography Fiber Clustering Framework for Functionally Consistent White Matter Parcellation Using Multimodal Diffusion MRI and Functional MRI

TL;DR

This work tackles the limitation of geometry-only fiber clustering by integrating diffusion MRI tractography with resting-state fMRI signals to produce functionally consistent white-matter parcellations. It introduces DMVFC, a two-stage deep multi-view clustering framework that uses two parallel DGCNNs to learn embeddings from geometry $\boldsymbol{x}_i^1 \in \mathbb{R}^{n_p \times 3}$ and function $\boldsymbol{x}_i^2 \in \mathbb{R}^{2 \times 600}$, with a self-supervised loss $L_s$ guiding the embeddings. Collaborative fine-tuning enforces cross-view agreement via $L_f = L_s + \gamma L_c$ with $\gamma = 0.1$, using cross-view KL divergences and distributions $P^v$ and $Q^v$ to harmonize geometry and function representations; inference averages per-view probabilities for final cluster assignments. On HCP-YA data, DMVFC achieves superior functional homogeneity and geometric coherence compared with QuickBundles and DFC, demonstrating that multimodal integration yields more meaningful WM parcellations and advancing tractography-based brain mapping.

Abstract

Tractography fiber clustering using diffusion MRI (dMRI) is a crucial strategy for white matter (WM) parcellation. Current methods primarily use the geometric information of fibers (i.e., the spatial trajectories) to group similar fibers into clusters, overlooking the important functional signals present along the fiber tracts. There is increasing evidence that neural activity in the WM can be measured using functional MRI (fMRI), offering potentially valuable multimodal information for fiber clustering. In this paper, we develop a novel deep learning fiber clustering framework, namely Deep Multi-view Fiber Clustering (DMVFC), that uses joint dMRI and fMRI data to enable functionally consistent WM parcellation. DMVFC can effectively integrate the geometric characteristics of the WM fibers with the fMRI BOLD signals along the fiber tracts. It includes two major components: 1) a multi-view pretraining module to compute embedding features from fiber geometric information and functional signals separately, and 2) a collaborative fine-tuning module to simultaneously refine the two kinds of embeddings. In the experiments, we compare DMVFC with two state-of-the-art fiber clustering methods and demonstrate superior performance in achieving functionally meaningful and consistent WM parcellation results.

Figures (3)

• Figure 1: Overview of DMVFC.
• Figure 2: Flowchart of data preprocessing.
• Figure 3: Example fiber clusters for the compared methods. The regions in red indicate a higher functional coherence, where the signals are similar across the fibers.