Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Q-space Guided Collaborative Attention Translation Network for Flexible Diffusion-Weighted Images Synthesis

Pengli Zhu, Yingji Fu, Nanguang Chen, Anqi Qiu

TL;DR

This work tackles the challenge of synthesizing high-angular-resolution diffusion data under flexible $q$-space sampling by introducing Q-CATN, a Q-space guided collaborative attention translation network. The model fuses multi-modal structural MRI through a SMA encoder, MMAF, and a $q$-space embedding with CBIN modulation, enabling dense, $q$-space-aware reconstruction via a SMA decoder. A $q$-space conditional discriminator provides adversarial feedback conditioned on the sampling coordinates, while specialized losses enforce data fidelity and anatomical consistency. Evaluated on the Human Connectome Project dataset, Q-CATN outperforms state-of-the-art methods in parameter-map estimation and fiber-tracking quality, demonstrating robustness to varied sampling schemes and potential for real-time clinical use due to improved efficiency.

Abstract

This study, we propose a novel Q-space Guided Collaborative Attention Translation Networks (Q-CATN) for multi-shell, high-angular resolution DWI (MS-HARDI) synthesis from flexible q-space sampling, leveraging the commonly acquired structural MRI data. Q-CATN employs a collaborative attention mechanism to effectively extract complementary information from multiple modalities and dynamically adjust its internal representations based on flexible q-space information, eliminating the need for fixed sampling schemes. Additionally, we introduce a range of task-specific constraints to preserve anatomical fidelity in DWI, enabling Q-CATN to accurately learn the intrinsic relationships between directional DWI signal distributions and q-space. Extensive experiments on the Human Connectome Project (HCP) dataset demonstrate that Q-CATN outperforms existing methods, including 1D-qDL, 2D-qDL, MESC-SD, and QGAN, in estimating parameter maps and fiber tracts both quantitatively and qualitatively, while preserving fine-grained details. Notably, its ability to accommodate flexible q-space sampling highlights its potential as a promising toolkit for clinical and research applications. Our code is available at https://github.com/Idea89560041/Q-CATN.

Q-space Guided Collaborative Attention Translation Network for Flexible Diffusion-Weighted Images Synthesis

TL;DR

This work tackles the challenge of synthesizing high-angular-resolution diffusion data under flexible -space sampling by introducing Q-CATN, a Q-space guided collaborative attention translation network. The model fuses multi-modal structural MRI through a SMA encoder, MMAF, and a -space embedding with CBIN modulation, enabling dense, -space-aware reconstruction via a SMA decoder. A -space conditional discriminator provides adversarial feedback conditioned on the sampling coordinates, while specialized losses enforce data fidelity and anatomical consistency. Evaluated on the Human Connectome Project dataset, Q-CATN outperforms state-of-the-art methods in parameter-map estimation and fiber-tracking quality, demonstrating robustness to varied sampling schemes and potential for real-time clinical use due to improved efficiency.

Abstract

This study, we propose a novel Q-space Guided Collaborative Attention Translation Networks (Q-CATN) for multi-shell, high-angular resolution DWI (MS-HARDI) synthesis from flexible q-space sampling, leveraging the commonly acquired structural MRI data. Q-CATN employs a collaborative attention mechanism to effectively extract complementary information from multiple modalities and dynamically adjust its internal representations based on flexible q-space information, eliminating the need for fixed sampling schemes. Additionally, we introduce a range of task-specific constraints to preserve anatomical fidelity in DWI, enabling Q-CATN to accurately learn the intrinsic relationships between directional DWI signal distributions and q-space. Extensive experiments on the Human Connectome Project (HCP) dataset demonstrate that Q-CATN outperforms existing methods, including 1D-qDL, 2D-qDL, MESC-SD, and QGAN, in estimating parameter maps and fiber tracts both quantitatively and qualitatively, while preserving fine-grained details. Notably, its ability to accommodate flexible q-space sampling highlights its potential as a promising toolkit for clinical and research applications. Our code is available at https://github.com/Idea89560041/Q-CATN.
Paper Structure (13 sections, 2 equations, 5 figures)

This paper contains 13 sections, 2 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overall Architecture
  4. Collaborative Attention Mechanism
  5. $Q$-space Embedding Module
  6. $Q$-space Conditional Discriminator
  7. Loss Function
  8. Implementation
  9. Experimental Results
  10. Qualitative Analysis
  11. Quantitative Analysis
  12. FOD and Tractography
  13. Conclusion

Figures (5)

  • Figure 1: Overview of Q-CATN. Panel (A) illustrates the high-level structure of the model. Panel (B) details the architecture of the single-modal attention module. Panel (C) shows the multi-modal attention fusion mechanism. Panel (D) outlines the structure of the conditional discriminator.
  • Figure 2: DWI synthesis results under different b-values configurations. On the left, all potential input channels (b0, T1, T2) are shown, while the right side displays the predicted results, with a standard DWI slice serving as the reference.
  • Figure 3: Qualitative comparison of different methods for undersampling DWI parameter fitting. The first six columns show the various diffusion maps and the last column is the ODI error map, each row represents different comparison methods, with the reference map at the bottom.
  • Figure 4: Quantitative comparison of estimated parameter maps in different methods.
  • Figure 5: Qualitative comparisons among different methods in fiber orientation distribution (FOD) and tractography. The top shows two magnified ROIs of FOD, and the bottom shows the tractography and two specific fiber tracts, i.e., Cingulum (L) and CorpusCallosum.