Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DiffuseReg: Denoising Diffusion Model for Obtaining Deformation Fields in Unsupervised Deformable Image Registration

Yongtai Zhuo, Yiqing Shen

TL;DR

DiffuseReg is introduced, an innovative diffusion-based method that denoises deformation fields instead of images for improved transparency and enhances control over the denoising registration process with a novel similarity consistency regularization.

Abstract

Deformable image registration aims to precisely align medical images from different modalities or times. Traditional deep learning methods, while effective, often lack interpretability, real-time observability and adjustment capacity during registration inference. Denoising diffusion models present an alternative by reformulating registration as iterative image denoising. However, existing diffusion registration approaches do not fully harness capabilities, neglecting the critical sampling phase that enables continuous observability during the inference. Hence, we introduce DiffuseReg, an innovative diffusion-based method that denoises deformation fields instead of images for improved transparency. We also propose a novel denoising network upon Swin Transformer, which better integrates moving and fixed images with diffusion time step throughout the denoising process. Furthermore, we enhance control over the denoising registration process with a novel similarity consistency regularization. Experiments on ACDC datasets demonstrate DiffuseReg outperforms existing diffusion registration methods by 1.32 in Dice score. The sampling process in DiffuseReg enables real-time output observability and adjustment unmatched by previous deep models.

DiffuseReg: Denoising Diffusion Model for Obtaining Deformation Fields in Unsupervised Deformable Image Registration

TL;DR

DiffuseReg is introduced, an innovative diffusion-based method that denoises deformation fields instead of images for improved transparency and enhances control over the denoising registration process with a novel similarity consistency regularization.

Abstract

Deformable image registration aims to precisely align medical images from different modalities or times. Traditional deep learning methods, while effective, often lack interpretability, real-time observability and adjustment capacity during registration inference. Denoising diffusion models present an alternative by reformulating registration as iterative image denoising. However, existing diffusion registration approaches do not fully harness capabilities, neglecting the critical sampling phase that enables continuous observability during the inference. Hence, we introduce DiffuseReg, an innovative diffusion-based method that denoises deformation fields instead of images for improved transparency. We also propose a novel denoising network upon Swin Transformer, which better integrates moving and fixed images with diffusion time step throughout the denoising process. Furthermore, we enhance control over the denoising registration process with a novel similarity consistency regularization. Experiments on ACDC datasets demonstrate DiffuseReg outperforms existing diffusion registration methods by 1.32 in Dice score. The sampling process in DiffuseReg enables real-time output observability and adjustment unmatched by previous deep models.
Paper Structure (19 sections, 4 equations, 3 figures, 2 tables)

This paper contains 19 sections, 4 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Problem Formulation
  4. Architecture for Denoising Network
  5. Condition Fusion in Denoising Network
  6. Masked MSA in a Window
  7. Attention Mask with Conditions
  8. Training Objective
  9. Consistency Reguarization
  10. Experiment
  11. Implementation Details
  12. Datasets and Metrics
  13. Results
  14. Ablation Study on Denoising Network
  15. Ablation Study on Regularization
  16. ...and 4 more sections

Figures (3)

  • Figure 1: (a) Traditional deep learning registration method, where a moving image is aligned to a fixed image using a learned deformation field generated by a registration network ('Reg'). (b) Existing approaches claiming to implement diffusion models for registration, shown here without the application of noise during the inference phase, which is signified by the 'Noisy' label next to the fixed image. (c) Our proposed registration with diffusion model. The looped arrow around the denoise network symbolizes the inclusion of a sampling process that iteratively denoises the deformation field.
  • Figure 2: (a) The overall workflow of the denoising network in DiffuseReg. (b) Swin Transformer block, showing conditional image feature integration. (c) The illustration of backbone utilizing conditional information. (d) Attention masking in DiffuseReg, with unmasked attention locations indicated by $1$ for selective feature focus.
  • Figure 3: The deformation fields and registered images during sampling process and the comparison with baseline methods. The images are annotated with masks of the cardiac tissue. The warped images within the blue and red boxes are overlaid with the intricate grid of deformation field.