Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FOD-Diff: 3D Multi-Channel Patch Diffusion Model for Fiber Orientation Distribution

Hao Tang, Hanyu Liu, Alessandro Perelli, Xi Chen, Chao Li

TL;DR

The paper tackles the challenge of estimating HAR-FOD from low-angular-resolution diffusion MRI by introducing FOD-Diff, a 3D multi-channel patch diffusion model. It combines a FOD-patch Adapter, a voxel-level Conditional Coordinating Module, and a Spherical Harmonic Attention mechanism to enable efficient, coherent HAR-FOD generation from LAR-FOD. Through patch-based diffusion with WM-guided sampling and SH-aware conditioning, the method achieves state-of-the-art HAR-FOD synthesis on HCP data, with ablations confirming the contribution of each component. This approach potentially reduces scan time while preserving fiber orientation accuracy, benefiting clinical diffusion MRI workflows and white-matter tractography analyses.

Abstract

Diffusion MRI (dMRI) is a critical non-invasive technique to estimate fiber orientation distribution (FOD) for characterizing white matter integrity. Estimating FOD from single-shell low angular resolution dMRI (LAR-FOD) is limited by accuracy, whereas estimating FOD from multi-shell high angular resolution dMRI (HAR-FOD) requires a long scanning time, which limits its applicability. Diffusion models have shown promise in estimating HAR-FOD based on LAR-FOD. However, using diffusion models to efficiently generate HAR-FOD is challenging due to the large number of spherical harmonic (SH) coefficients in FOD. Here, we propose a 3D multi-channel patch diffusion model to predict HAR-FOD from LAR-FOD. We design the FOD-patch adapter by introducing the prior brain anatomy for more efficient patch-based learning. Furthermore, we introduce a voxel-level conditional coordinating module to enhance the global understanding of the model. We design the SH attention module to effectively learn the complex correlations of the SH coefficients. Our experimental results show that our method achieves the best performance in HAR-FOD prediction and outperforms other state-of-the-art methods.

FOD-Diff: 3D Multi-Channel Patch Diffusion Model for Fiber Orientation Distribution

TL;DR

The paper tackles the challenge of estimating HAR-FOD from low-angular-resolution diffusion MRI by introducing FOD-Diff, a 3D multi-channel patch diffusion model. It combines a FOD-patch Adapter, a voxel-level Conditional Coordinating Module, and a Spherical Harmonic Attention mechanism to enable efficient, coherent HAR-FOD generation from LAR-FOD. Through patch-based diffusion with WM-guided sampling and SH-aware conditioning, the method achieves state-of-the-art HAR-FOD synthesis on HCP data, with ablations confirming the contribution of each component. This approach potentially reduces scan time while preserving fiber orientation accuracy, benefiting clinical diffusion MRI workflows and white-matter tractography analyses.

Abstract

Diffusion MRI (dMRI) is a critical non-invasive technique to estimate fiber orientation distribution (FOD) for characterizing white matter integrity. Estimating FOD from single-shell low angular resolution dMRI (LAR-FOD) is limited by accuracy, whereas estimating FOD from multi-shell high angular resolution dMRI (HAR-FOD) requires a long scanning time, which limits its applicability. Diffusion models have shown promise in estimating HAR-FOD based on LAR-FOD. However, using diffusion models to efficiently generate HAR-FOD is challenging due to the large number of spherical harmonic (SH) coefficients in FOD. Here, we propose a 3D multi-channel patch diffusion model to predict HAR-FOD from LAR-FOD. We design the FOD-patch adapter by introducing the prior brain anatomy for more efficient patch-based learning. Furthermore, we introduce a voxel-level conditional coordinating module to enhance the global understanding of the model. We design the SH attention module to effectively learn the complex correlations of the SH coefficients. Our experimental results show that our method achieves the best performance in HAR-FOD prediction and outperforms other state-of-the-art methods.

Paper Structure

This paper contains 14 sections, 7 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview.
  4. FOD-patch Adapter
  5. Voxel-level Conditional Coordinating Module
  6. Spherical Harmonic Attention Module
  7. Experiments and Results
  8. Datasets and Preprocessing
  9. Implementation Details
  10. Comparisons with SOTA and baselines
  11. Quantitative Results.
  12. Qualitative Results.
  13. Ablation Experiments
  14. Conclusion

Figures (2)

  • Figure 1: Model architecture. FPA provides the location of optimal patch $p_{k}$ by probabilistic sampling and fuses multi-scale patch feature for guidance. In the forward process, the HAR-FOD patch $S_{0}$ is gradually noised by $q(S_{t}|S_{t-1})$. In the reverse process, the condition patches are obtained by CCM and FPA from LAR-FOD to guide training. The condition patches and noised HAR-FOD patch $S_{t}$ are concatenated to train U-Net for $p_{\theta}(S_{t-1}|S_{t}, c)$. SHAM replaces the traditional output layer of U-Net to predict the forward noise patterns more effectively. In inference, predicted patches $S^{r}_{0}$ are merged by post-processing operations, and the predicted HAR-FOD is obtained.
  • Figure 2: Selected samples and their regions of interest (ROIs) for comparison. ROI is selected in the corpus callosum region and a typical cross fiber is zoomed in.