List of Papers Browse by Subject Areas Author List
Abstract
The deformable registration of images of different modalities, essential in many medical imaging applications, remains challenging. The main challenge is developing a robust measure for image overlap despite the compared images capturing different aspects of the underlying tissue. Here, we explore similarity metrics based on functional dependence between intensity values of registered images. Although functional dependence is too restrictive on the global scale, earlier work has shown competitive performance in deformable registration when such measures are applied over small enough contexts. We confirm this finding and further develop the idea by modeling local functional dependence via the linear basis function model with the basis functions learned jointly with the deformation. The measure can be implemented via convolutions, making it efficient to compute on GPUs. We release the method as an easy-to-use tool and show good performance on three datasets compared to well-established baseline and earlier functional dependence-based methods.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1747_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/honkamj/locor
https://github.com/honkamj/locor-experiments
Link to the Dataset(s)
Learn2Reg 2021 Task 1 data: https://learn2reg.grand-challenge.org/Learn2Reg2021/
CERMEP-IDB-MRXFDG dataset: https://pubmed.ncbi.nlm.nih.gov/34529159/
IXI dataset: http://brain-development.org/ixi-dataset/
BibTex
@InProceedings{HonJoe_New_MICCAI2025,
author = { Honkamaa, Joel and Marttinen, Pekka},
title = { { New Multimodal Similarity Measure for Image Registration via Modeling Local Functional Dependence with Linear Combination of Learned Basis Functions } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15961},
month = {September},
page = {397 -- 407}
}
Reviews
Review #1
- Please describe the contribution of the paper
The main contribution of the paper is a novel similarity measure for multimodal deformable (medical) image registration. This similarity measure models local intensity relationships via linear combinations of learned basis functions as opposed to Parzen window approximations. In doing so, the method can be implemented with a convolution operation, and is easily parallelizable via gpu.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The major strengths of the paper lie in the mathematically sound and interesting methodology of modeling the local functional dependency between image intensities using a linear combination of learned bases. This is a strength from two perspectives: (1) because the method doesn’t rely on fixed polynomial bases or kde the approach is perhaps more flexible and potentially more efficient, (2) the approach aggregates local losses with respect to local neighborhoods, allowing for the deformation to be learned efficiently but also mitigating the sensitivity to spatial changes in relationships between image intensities that would arise if using a purely global measure. While other learned deformation methods exist, this particular approach has not been thoroughly explored before (to the best of my knowledge).
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
While the methodology itself is interesting, its empirical usefulness in the medical imaging space specifically is a bit lacking. The authors consider 3 datasets, 2 are real MRI-CT registration and the third is a semi-synthetic dataset for head MRI T2-PDregistration. Several aspects of the experimental setup, data, and eval pipeline could be improved upon.
The number of samples is very limited. 8 pairs for the first dataset and 37 pairs for the second MRI-CT is not enough to really understand the strengths and weaknesses of the method itself. Running studies on datasets with more paired samples (or simply additional datasets of different anatomical regions would be more convincing). The authors even mention that the 8 sample dataset is too small to draw statistically significant conclusions. Agreed, but this is fixable with more experiments on more data.
Multimodalities considered: The experiment section could benefit from considering a more diverse range of modality pairings to really probe how well the method performs against other approaches (i.e. CT-PET, MRI-PET, MR-Ultrasounds etc)
In addition to DICE, MAE, and TRE, the experiment section could benefit from considering downstream clinically relevant metrics. There is a potential gap between registration accuracy metrics and clinical utility (i.e. does the registration sufficiently improve diagnostic accuracy, procedure planning, or therapeutic outcomes). Considering a clinical application area and evaluating the method from that perspective would be a useful addition.
It would be nice to discuss the limitations of the model more in depth. In what situations (clinical) does the methodology excel or fail at.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission has provided an anonymized link to the source code, dataset, or any other dependencies.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The methodology seems to be novel and sound, but the experiment section can be improved upon. See weaknesses for suggestions on how to improve.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
This paper proposes Locor, a new similarity measure for multimodal image registration. It models local functional dependence between modalities as a linear combination of learned basis functions (parametrized by small MLPs). The similarity can be computed efficiently on GPUs via convolutions. Locor is evaluated on three tasks: intra-patient Abdomen MRI-CT, head MRI-CT, and head MRI T2-PD.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
- Locor captures spatially varying intensity relationships and outperforms existing multimodal similarity measures on three registration tasks.
- The basis functions are learned jointly with the deformation, providing more flexibility than fixed affine or polynomial functions.
- Spatial derivative is further used to improve the results
- The similarity loss can be computed efficiently with convolutions and is integrated into a multi-resolution registration pipeline
- An extensive hyperparameter optimization on the validation set is conducted.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- Fig. 1 is informative to show the advantage of learned basis function over polynomial functions. But this is the only 1 visual results in the paper…
- The linear combination of learned nonlinear basis functions (MLPs) could itself be viewed as a global function. What is the rationale behind learning the MLPs globally but solving for the local coefficients via linear least squares? Could the entire function be learned jointly instead?
- Further clarification is needed on the joint optimization of local functions and registration. When images are poorly pre-aligned, could the local function overfit (as noted: “having too high number of basis functions can be detrimental”)? How is this issue mitigated in practice?
- The method uses bidirectional registration and bending energy regularization. Is there any synergy or interaction between the proposed similarity measure and the regularization/parametrization of the deformation?
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
- The learned local dependence can be interpreted as modality translation from image B to image A to some extent (correct me if I was wrong). Visualization of the function output f(IB) over optimization could be informative and interesting.
- A simple pseudo-code would be helpful to understand the nested optimization of the basis function, linear coefficients and registration. But I also understand that the space is limited.
- It would be interesting to analyze the effect of initial misalignment and the number of basis functions. What are the typical failure cases?
- A study on the interaction between the similarity measure and deformation regularization would provide further insight.
- What is the rationale behind the spatial derivates? Does the edge information extracted help the basis function learning?
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The proposed method is simple, well-motivated, and effective across three multimodal registration tasks. While more qualitative results could be added, some methodological details could be clarified, and additional discussion would strengthen the paper, I believe it aligns well with MICCAI’s scope and interests.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
Local (neighbourhood) functional dependence similarity metric for multi-modal, deformable image registration.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
Demonstration that learned (fully connected network) basis functions, for the functional dependence between modalities, improves registration performance.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
Additional optimisation complexity of learning neural network weights for basis functions.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission has provided an anonymized link to the source code, dataset, or any other dependencies.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2025/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Valuable contribution of development of local (neighbourhood) functional dependence similarity metric for multi-modal, deformable image registration.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
N/A
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A