A Unified Learning Model for Estimating Fiber Orientation Distribution Functions on Heterogeneous Multi-shell Diffusion-weighted MRI

TL;DR

This work tackles estimating fiber orientation distribution functions (fODFs) from heterogeneous multi-shell DW-MRI by proposing a unified, single-stage dynamic network built on spherical convolutional neural networks. The model uses a dynamic head conditioned on the available shell configurations to handle $2^K-1$ possible multi-shell setups, enabling end-to-end estimation of fODFs and tissue fractions without retraining for each shell configuration. Evaluated on the HCP-ya dataset, the approach outperforms traditional multi-stage pipelines in repeat fODF estimation across shell dropoff and single-shell sequences, while matching or approaching the performance of shell-specific models. The key contributions include (1) a unified dynamic network capable of learning across arbitrary shell configurations, (2) a spherical CNN architecture that explicitly accounts for gradient schemes, and (3) demonstrated superior robustness and efficiency compared with MSMT-CSD and SHORE baselines, with implications for faster, more generalizable diffusion MRI analysis.

Abstract

Diffusion-weighted (DW) MRI measures the direction and scale of the local diffusion process in every voxel through its spectrum in q-space, typically acquired in one or more shells. Recent developments in micro-structure imaging and multi-tissue decomposition have sparked renewed attention to the radial b-value dependence of the signal. Applications in tissue classification and micro-architecture estimation, therefore, require a signal representation that extends over the radial as well as angular domain. Multiple approaches have been proposed that can model the non-linear relationship between the DW-MRI signal and biological microstructure. In the past few years, many deep learning-based methods have been developed towards faster inference speed and higher inter-scan consistency compared with traditional model-based methods (e.g., multi-shell multi-tissue constrained spherical deconvolution). However, a multi-stage learning strategy is typically required since the learning process relies on various middle representations, such as simple harmonic oscillator reconstruction (SHORE) representation. In this work, we present a unified dynamic network with a single-stage spherical convolutional neural network, which allows efficient fiber orientation distribution function (fODF) estimation through heterogeneous multi-shell diffusion MRI sequences. We study the Human Connectome Project (HCP) young adults with test-retest scans. From the experimental results, the proposed single-stage method outperforms prior multi-stage approaches in repeated fODF estimation with shell dropoff and single-shell DW-MRI sequences.

Figures (2)

• Figure 1: Utilizing multi-shell DW-MRI signals in deep learning usually requires independent models trained for each specific shell configuration as conventional SH-based modeling cannot directly leverage additional information (radial space) provided by multi-shell acquisitions. In our study, the dynamic head aims to improve the network expressiveness by learning and adaptively adjusting the first convolution layer for different shell configurations.
• Figure 2: This is a visualization of the fODF prediction and the correlation with the GT in different views. The background of the zoom-in patches shows the ACC spatial map with the GT signals.