A Novel Dual-Stream Framework for dMRI Tractography Streamline Classification with Joint dMRI and fMRI Data

TL;DR

This work tackles the challenge of distinguishing functionally distinct subdivisions within diffusion MRI tractography by integrating fMRI data at tract endpoints with full geometric trajectory information. It introduces a dual-stream architecture comprising a frozen geometric backbone and a trainable auxiliary pathway that processes endpoint fMRI signals, fused at the logit level via $logits_{final} = logits_{backbone} + logits_{auxiliary}$. Evaluated on Human Connectome Project data, the method demonstrates accurate CST somatotopy (leg, trunk, face, hand) with state-of-the-art performance (F1 around 0.90), outperforming single-modality and other multimodal baselines. The approach offers robust, functionally informed tract parcellation and has potential to generalize to other white matter pathways, enhancing neuroscience research and clinical planning.

Abstract

Streamline classification is essential to identify anatomically meaningful white matter tracts from diffusion MRI (dMRI) tractography. However, current streamline classification methods rely primarily on the geometric features of the streamline trajectory, failing to distinguish between functionally distinct fiber tracts with similar pathways. To address this, we introduce a novel dual-stream streamline classification framework that jointly analyzes dMRI and functional MRI (fMRI) data to enhance the functional coherence of tract parcellation. We design a novel network that performs streamline classification using a pretrained backbone model for full streamline trajectories, while augmenting with an auxiliary network that processes fMRI signals from fiber endpoint regions. We demonstrate our method by parcellating the corticospinal tract (CST) into its four somatotopic subdivisions. Experimental results from ablation studies and comparisons with state-of-the-art methods demonstrate our approach's superior performance.

Figures (2)

• Figure 1: Overview of our model pipeline
• Figure 2: The top row is the CST region colored by PCA-reduced fMRI endpoint signals, and the bottom row is the comparison of the prediction results for 'trunk' class from the geometry-only model and the proposed model against its ground truth region.