Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MemWarp: Discontinuity-Preserving Cardiac Registration with Memorized Anatomical Filters

Hang Zhang, Xiang Chen, Renjiu Hu, Dongdong Liu, Gaolei Li, Rongguang Wang

TL;DR

Cardiac deformable registration is hindered by local discontinuities at organ boundaries, which violate global smoothness priors. MemWarp integrates a Laplacian pyramid warping network (LapWarp) with a memory network that produces region-specific dynamic filters guided by fixed-image features, enabling discontinuity-preserving registration without masks at inference. The method optimizes a composite loss incorporating similarity, dissimilarity, a gradient regularizer, and a region-based memory loss, and is trained to enforce region-aware discontinuities through deep supervision. On the public ACDC cardiac dataset, MemWarp outperforms state-of-the-art semi-/weakly-supervised baselines, achieving a Dice improvement of about 7.1 percentage points and producing realistic deformations suitable for downstream analysis and segmentation.

Abstract

Many existing learning-based deformable image registration methods impose constraints on deformation fields to ensure they are globally smooth and continuous. However, this assumption does not hold in cardiac image registration, where different anatomical regions exhibit asymmetric motions during respiration and movements due to sliding organs within the chest. Consequently, such global constraints fail to accommodate local discontinuities across organ boundaries, potentially resulting in erroneous and unrealistic displacement fields. In this paper, we address this issue with MemWarp, a learning framework that leverages a memory network to store prototypical information tailored to different anatomical regions. MemWarp is different from earlier approaches in two main aspects: firstly, by decoupling feature extraction from similarity matching in moving and fixed images, it facilitates more effective utilization of feature maps; secondly, despite its capability to preserve discontinuities, it eliminates the need for segmentation masks during model inference. In experiments on a publicly available cardiac dataset, our method achieves considerable improvements in registration accuracy and producing realistic deformations, outperforming state-of-the-art methods with a remarkable 7.1\% Dice score improvement over the runner-up semi-supervised method. Source code will be available at https://github.com/tinymilky/Mem-Warp.

MemWarp: Discontinuity-Preserving Cardiac Registration with Memorized Anatomical Filters

TL;DR

Cardiac deformable registration is hindered by local discontinuities at organ boundaries, which violate global smoothness priors. MemWarp integrates a Laplacian pyramid warping network (LapWarp) with a memory network that produces region-specific dynamic filters guided by fixed-image features, enabling discontinuity-preserving registration without masks at inference. The method optimizes a composite loss incorporating similarity, dissimilarity, a gradient regularizer, and a region-based memory loss, and is trained to enforce region-aware discontinuities through deep supervision. On the public ACDC cardiac dataset, MemWarp outperforms state-of-the-art semi-/weakly-supervised baselines, achieving a Dice improvement of about 7.1 percentage points and producing realistic deformations suitable for downstream analysis and segmentation.

Abstract

Many existing learning-based deformable image registration methods impose constraints on deformation fields to ensure they are globally smooth and continuous. However, this assumption does not hold in cardiac image registration, where different anatomical regions exhibit asymmetric motions during respiration and movements due to sliding organs within the chest. Consequently, such global constraints fail to accommodate local discontinuities across organ boundaries, potentially resulting in erroneous and unrealistic displacement fields. In this paper, we address this issue with MemWarp, a learning framework that leverages a memory network to store prototypical information tailored to different anatomical regions. MemWarp is different from earlier approaches in two main aspects: firstly, by decoupling feature extraction from similarity matching in moving and fixed images, it facilitates more effective utilization of feature maps; secondly, despite its capability to preserve discontinuities, it eliminates the need for segmentation masks during model inference. In experiments on a publicly available cardiac dataset, our method achieves considerable improvements in registration accuracy and producing realistic deformations, outperforming state-of-the-art methods with a remarkable 7.1\% Dice score improvement over the runner-up semi-supervised method. Source code will be available at https://github.com/tinymilky/Mem-Warp.
Paper Structure (11 sections, 4 equations, 2 figures, 2 tables)

This paper contains 11 sections, 4 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Preliminaries
  4. Laplacian Pyramid Warping Network
  5. Discontinuity-Preserving Deformable Registration
  6. Loss function & Overall Framework
  7. Experiments & Results
  8. Implementation Details & Comparator Methods
  9. Results & Analysis
  10. Discussions
  11. Conclusions

Figures (2)

  • Figure 1: Schematic representation of the MemWarp framework. The left panel depicts a 2-level LapWarp network employing Laplacian image pyramids; the right panel outlines the operation of the memory network.
  • Figure 2: Comparative visualization of MemWarp against other methods on cardiac MR images, highlighting deformable registration across ED $\Leftrightarrow$ ES phases. Pink arrows show omitted trabeculations; orange arrows identify artifacts. The right panel focuses on deformation fields outlined by the left panel's yellow dash, with arrow darkness indicating displacement magnitude.