Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

QID$^2$: An Image-Conditioned Diffusion Model for Q-space Up-sampling of DWI Data

Zijian Chen, Jueqi Wang, Archana Venkataraman

TL;DR

This study highlights the potential of diffusion models, and QID2 in particular, for q-space up-sampling, thus offering a promising toolkit for clinical and research applications.

Abstract

We propose an image-conditioned diffusion model to estimate high angular resolution diffusion weighted imaging (DWI) from a low angular resolution acquisition. Our model, which we call QID$^2$, takes as input a set of low angular resolution DWI data and uses this information to estimate the DWI data associated with a target gradient direction. We leverage a U-Net architecture with cross-attention to preserve the positional information of the reference images, further guiding the target image generation. We train and evaluate QID$^2$ on single-shell DWI samples curated from the Human Connectome Project (HCP) dataset. Specifically, we sub-sample the HCP gradient directions to produce low angular resolution DWI data and train QID$^2$ to reconstruct the missing high angular resolution samples. We compare QID$^2$ with two state-of-the-art GAN models. Our results demonstrate that QID$^2$ not only achieves higher-quality generated images, but it consistently outperforms the GAN models in downstream tensor estimation across multiple metrics. Taken together, this study highlights the potential of diffusion models, and QID$^2$ in particular, for q-space up-sampling, thus offering a promising toolkit for clinical and research applications.

QID$^2$: An Image-Conditioned Diffusion Model for Q-space Up-sampling of DWI Data

TL;DR

This study highlights the potential of diffusion models, and QID2 in particular, for q-space up-sampling, thus offering a promising toolkit for clinical and research applications.

Abstract

We propose an image-conditioned diffusion model to estimate high angular resolution diffusion weighted imaging (DWI) from a low angular resolution acquisition. Our model, which we call QID, takes as input a set of low angular resolution DWI data and uses this information to estimate the DWI data associated with a target gradient direction. We leverage a U-Net architecture with cross-attention to preserve the positional information of the reference images, further guiding the target image generation. We train and evaluate QID on single-shell DWI samples curated from the Human Connectome Project (HCP) dataset. Specifically, we sub-sample the HCP gradient directions to produce low angular resolution DWI data and train QID to reconstruct the missing high angular resolution samples. We compare QID with two state-of-the-art GAN models. Our results demonstrate that QID not only achieves higher-quality generated images, but it consistently outperforms the GAN models in downstream tensor estimation across multiple metrics. Taken together, this study highlights the potential of diffusion models, and QID in particular, for q-space up-sampling, thus offering a promising toolkit for clinical and research applications.
Paper Structure (7 sections, 9 equations, 3 figures, 1 table)

This paper contains 7 sections, 9 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. A Diffusion Model for Q-space Up-sampling of DWI
  4. Baseline Comparison Methods
  5. Implementation Details
  6. Experimental Results
  7. Conclusion

Figures (3)

  • Figure 1: QID$^2$ framework for up-sampling the angular resolution of DWI. The gray sphere represents the q-space. Red marks are the directions in the low angular resolution scan, and blue marks are the target gradient directions for image generation.
  • Figure 2: Qualitative results that compare the ground-truth DWI acquisition to images generated by QID$^2$ and the baselines methods for $R=3$ and $R=6$. Zoomed-in area highlights details that are preserved by our method and do not appear in the baselines.
  • Figure 3: Fiber direction and FA value map estimated based on the ground truth images, QID$^2$, and baseline methods. Row 1/3: Colored FA maps indicating fiber orientation, with minimal visually detectable differences among the images. Row 2/4: FA value maps, where brighter colors indicate higher FA values. Significant differences compared to the ground truth are zoomed-out with orange boxes.