Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large Scale Unsupervised Brain MRI Image Registration Solution for Learn2Reg 2024

Yuxi Zhang, Xiang Chen, Jiazheng Wang, Min Liu, Yaonan Wang, Dongdong Liu, Renjiu Hu, Hang Zhang

TL;DR

This work tackles unsupervised inter-subject brain MRI registration on a large-scale dataset (Learn2Reg Task 2 LUMIR) by designing an efficient backbone that cascades deformation refinement and incorporates Co-Attention, Large Kernel convolutions, Bilateral Filtering, and expanded channels. The model is trained with a composite loss combining a similarity term via NCC and a smoothness regularizer, enabling robust deformation fields without segmentation labels. On the OpenBHB dataset, it achieves a Dice of $77.34\%$, a $1.4\%$ improvement over TransMorph, along with reductions in non-diffeomorphic volumes and HD95, and ablation studies highlight contributions from BF, LK, CA, and LC modules. These results underscore backbone efficiency as a key driver for scalable, accurate brain MRI registration in large datasets, with practical implications for fast, unsupervised inter-subject alignment.

Abstract

In this paper, we summarize the methods and experimental results we proposed for Task 2 in the learn2reg 2024 Challenge. This task focuses on unsupervised registration of anatomical structures in brain MRI images between different patients. The difficulty lies in: (1) without segmentation labels, and (2) a large amount of data. To address these challenges, we built an efficient backbone network and explored several schemes to further enhance registration accuracy. Under the guidance of the NCC loss function and smoothness regularization loss function, we obtained a smooth and reasonable deformation field. According to the leaderboard, our method achieved a Dice coefficient of 77.34%, which is 1.4% higher than the TransMorph. Overall, we won second place on the leaderboard for Task 2.

Large Scale Unsupervised Brain MRI Image Registration Solution for Learn2Reg 2024

TL;DR

This work tackles unsupervised inter-subject brain MRI registration on a large-scale dataset (Learn2Reg Task 2 LUMIR) by designing an efficient backbone that cascades deformation refinement and incorporates Co-Attention, Large Kernel convolutions, Bilateral Filtering, and expanded channels. The model is trained with a composite loss combining a similarity term via NCC and a smoothness regularizer, enabling robust deformation fields without segmentation labels. On the OpenBHB dataset, it achieves a Dice of , a improvement over TransMorph, along with reductions in non-diffeomorphic volumes and HD95, and ablation studies highlight contributions from BF, LK, CA, and LC modules. These results underscore backbone efficiency as a key driver for scalable, accurate brain MRI registration in large datasets, with practical implications for fast, unsupervised inter-subject alignment.

Abstract

In this paper, we summarize the methods and experimental results we proposed for Task 2 in the learn2reg 2024 Challenge. This task focuses on unsupervised registration of anatomical structures in brain MRI images between different patients. The difficulty lies in: (1) without segmentation labels, and (2) a large amount of data. To address these challenges, we built an efficient backbone network and explored several schemes to further enhance registration accuracy. Under the guidance of the NCC loss function and smoothness regularization loss function, we obtained a smooth and reasonable deformation field. According to the leaderboard, our method achieved a Dice coefficient of 77.34%, which is 1.4% higher than the TransMorph. Overall, we won second place on the leaderboard for Task 2.
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. Methodology
  3. Proposed Method
  4. Loss Function
  5. Total Loss.
  6. NCC Loss.
  7. Regularization Loss.
  8. Experimental Setup and Results
  9. Datasets
  10. Experimental Setup
  11. Results
  12. Analysis and Conclusion
  13. Acknowledgments.

Figures (3)

  • Figure 1: The backbone network includes a general convolutional block and cascade registration structure. To enhance its efficiency, several schemes/structures are incorporated as shown at the bottom of the figure.
  • Figure 2: The loss curve and dice coefficients on validation set during the training process
  • Figure 3: Visualization of three test samples in the submitted data. From left to right are the moving image, fixed image, predicted motion vector, and predicted deformation field.