Symmetric Multi-level Gradient-Inverse Consistency Network for Brain Image Registration with Large Deformation

HaoyingBai, Tongtong Che, JichangZhang, Shuyu Li

Keywords: Symmetric registration, Consistency-Constrained, Inverse-Consistent, Multi-level

Abstract:

Accurate and robust deformable image registration is crucial for brain image analysis. While deep learning has significantly advanced this field, existing methods often lack robustness for large deformations due to inter-subject variability, frequently requiring pre-registration and relying heavily on data-driven approaches. To address these limitations, we propose an end-to-end Symmetric Multis-level Gradient-Inverse Consistency Network (SM-GICNet) for accurate and robust brain image registration. SM-GICNet employs 1) a symmetric multi-level framework with an attention gate mechanism to capture complex deformations at multiple scales, 2) a symmetric registration strategy at each level to mitigate directional bias, and 3) a gradient inverse consistency strategy to reduce reliance on data-driven constraints and control deformation field complexity. Experimental results demonstrate that our method is able to eliminate the need for pre-registration and
outperforms state-of-the-art methods on large deformation registration tasks, achieving a Dice similarity coefficient of 0.797. The implementation of our SM-GICNet is available online at https://github.com/LSYLAB/SM-GICNet.git.

Primary Subject Area: Image Registration

Secondary Subject Area: Application: Neuroimaging

Paper Type: Both

Registration Requirement: Yes

Reproducibility: https://github.com/LSYLAB/SM-GICNet.git

Visa & Travel: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Created a single midl25_NNN.zip file with midl25_NNN.tex, midl25_NNN.bib, all necessary figures and files., Includes \documentclass{midl}, \jmlryear{2025}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package, Did not use the times package., Author and institution details are de-anonymized where needed. All author names, affiliations, and paper title are correctly spelled and capitalized in the biography section., References must use the .bib file. Did not override the bibliographystyle defined in midl.cls. Did not use \begin{thebibliography} directly to insert references., Tables and figures do not overflow margins; avoid using \scalebox; used \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., Appendices and supplementary material are included in the same PDF after references., Main paper does not exceed 9 pages; acknowledgements, references, and appendix start on page 10 or later.

Submission Number: 147