Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MrRegNet: Multi-resolution Mask Guided Convolutional Neural Network for Medical Image Registration with Large Deformations

Ruizhe Li, Grazziela Figueredo, Dorothee Auer, Christian Wagner, Xin Chen

TL;DR

Deformable medical image registration with large deformations is challenging, especially achieving accurate local alignment. MrRegNet addresses this with a single-encoder, multi-resolution CNN that estimates residual displacement fields $D_i$ across $K$ levels and fuses them to produce a diffeomorphic deformation, guided by a mask-based loss to emphasize ROI alignment. In experiments on the OASIS 3D brain MRI dataset and a local 2D brain MRI dataset, MrRegNet outperforms Demons and VoxelMorph in global similarity while the mask-guided variant MrRegNet-M delivers notable gains in local region metrics (DSC, HD) without sacrificing global performance. The approach offers improved local registration accuracy, efficiency through a lightweight architecture, and potential segmentation benefits via accurate mask warping.

Abstract

Deformable image registration (alignment) is highly sought after in numerous clinical applications, such as computer aided diagnosis and disease progression analysis. Deep Convolutional Neural Network (DCNN)-based image registration methods have demonstrated advantages in terms of registration accuracy and computational speed. However, while most methods excel at global alignment, they often perform worse in aligning local regions. To address this challenge, this paper proposes a mask-guided encoder-decoder DCNN-based image registration method, named as MrRegNet. This approach employs a multi-resolution encoder for feature extraction and subsequently estimates multi-resolution displacement fields in the decoder to handle the substantial deformation of images. Furthermore, segmentation masks are employed to direct the model's attention toward aligning local regions. The results show that the proposed method outperforms traditional methods like Demons and a well-known deep learning method, VoxelMorph, on a public 3D brain MRI dataset (OASIS) and a local 2D brain MRI dataset with large deformations. Importantly, the image alignment accuracies are significantly improved at local regions guided by segmentation masks. Github link:https://github.com/ruizhe-l/MrRegNet.

MrRegNet: Multi-resolution Mask Guided Convolutional Neural Network for Medical Image Registration with Large Deformations

TL;DR

Deformable medical image registration with large deformations is challenging, especially achieving accurate local alignment. MrRegNet addresses this with a single-encoder, multi-resolution CNN that estimates residual displacement fields across levels and fuses them to produce a diffeomorphic deformation, guided by a mask-based loss to emphasize ROI alignment. In experiments on the OASIS 3D brain MRI dataset and a local 2D brain MRI dataset, MrRegNet outperforms Demons and VoxelMorph in global similarity while the mask-guided variant MrRegNet-M delivers notable gains in local region metrics (DSC, HD) without sacrificing global performance. The approach offers improved local registration accuracy, efficiency through a lightweight architecture, and potential segmentation benefits via accurate mask warping.

Abstract

Deformable image registration (alignment) is highly sought after in numerous clinical applications, such as computer aided diagnosis and disease progression analysis. Deep Convolutional Neural Network (DCNN)-based image registration methods have demonstrated advantages in terms of registration accuracy and computational speed. However, while most methods excel at global alignment, they often perform worse in aligning local regions. To address this challenge, this paper proposes a mask-guided encoder-decoder DCNN-based image registration method, named as MrRegNet. This approach employs a multi-resolution encoder for feature extraction and subsequently estimates multi-resolution displacement fields in the decoder to handle the substantial deformation of images. Furthermore, segmentation masks are employed to direct the model's attention toward aligning local regions. The results show that the proposed method outperforms traditional methods like Demons and a well-known deep learning method, VoxelMorph, on a public 3D brain MRI dataset (OASIS) and a local 2D brain MRI dataset with large deformations. Importantly, the image alignment accuracies are significantly improved at local regions guided by segmentation masks. Github link:https://github.com/ruizhe-l/MrRegNet.
Paper Structure (13 sections, 3 equations, 3 figures, 2 tables)

This paper contains 13 sections, 3 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Model Architecture
  4. Model Training
  5. Method Evaluation
  6. Materials and Experiments
  7. Parameter Settings
  8. Results
  9. 2D Brain MRI Data
  10. 3D Brain MRI Data
  11. Discussion and Conclusions
  12. Compliance with Ethical Standards
  13. Conflicts of Interest

Figures (3)

  • Figure 1: Overview of the proposed multi-resolution image registration framework with 3 levels.
  • Figure 2: Visualisation results of different registration methods without mask guided loss term on the 2D brain dataset.
  • Figure 3: The visualization of MrRegNet with and without mask guided loss.