Efficient Adaptive Filtering for Deformable Image registration
Keywords: Deformable image registration, Adaptive filtering, Bilateral Grid, Piece-wise Smooth
TL;DR: A novel efficient model for medical image registration using differentiable bilateral grid
Abstract: In medical image registration, where targets exhibit piecewise smooth structures, a carefully designed low-resolution data structure can effectively approximate full-resolution deformation fields with minimal accuracy loss.
Although this physical prior has proven effective in traditional registration algorithms, it remains underexplored in current learning-based registration literature.
In this paper, we propose AdaWarp, a novel neural network module that leverages this prior for efficient and accurate medical image registration.
AdaWarp comprises an encoder, a guidance map generator, and a differentiable bilateral grid, enabling an edge-preserving low-frequency approximation of the deformation field.
This design reduces computational complexity with low-resolution feature maps while increasing the effective receptive field, achieving a balanced trade-off between registration accuracy and efficiency.
Experiments on two registration datasets covering different modalities and input constraints demonstrate that AdaWarp outperforms existing methods in accuracy-efficiency and accuracy-smoothness tradeoffs.
Primary Area: interpretability and explainable AI
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 13590
Loading