A Plug-In Curriculum Scheduler for Improved Deformable Medical Image Registration
Keywords: medical image registration; curriculum learning
Abstract: Deformable image registration is a crucial task in medical image analysis, and its complexity has spurred significant research and ongoing progress.
Much of the work in this area has concentrated on achieving incremental performance gains by adjusting network architectures or introducing new loss functions.
However, these modifications are often tailored to specific tasks or datasets, which limits their general applicability.
To address this limitation, we propose an innovative solution: a plug-in curriculum scheduler that can be seamlessly integrated into existing methods without changing their core architecture.
Our scheduler, inspired by curriculum learning, progressively increases task difficulty to enhance performance, incorporating sample difficulty and matching accuracy as key criteria.
Sample difficulty is assessed at voxel and volume levels, using Variance of Gradients for voxel complexity and Gaussian blurring for volume evaluation, while matching accuracy involves gradually increasing supervision for improved alignment and accuracy.
We empirically demonstrate that this scheduler achieves superior accuracy and visual quality in various tasks and datasets.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
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: 1138
Loading