List of Papers Browse by Subject Areas Author List
Abstract
Designing generative models for 3D structural brain MRI that synthesize morphologically-plausible and attribute-specific (e.g., age, sex, disease state) samples is an active area of research. Existing approaches based on frameworks like GANs or diffusion models synthesize the image directly, which may limit their ability to capture intricate morphological details. In this work, we propose a 3D brain MRI generation method based on state-of-the-art latent diffusion models (LDMs), called MorphLDM, that generates novel images by synthesizing deformation fields applied to a learned template. Instead of using a reconstruction-based autoencoder (as in LDM), our encoder outputs a latent embedding derived from both an image and a learned template that is itself the output of a template decoder; this latent is passed to a deformation field decoder, whose output is applied to the learned template. A registration loss is minimized between the original image and the deformed template with respect to the encoder and both decoders. Empirically, our approach outperforms generative baselines on metrics spanning image diversity, adherence with respect to input conditions, and voxel-based morphometry. Our code will be made available upon acceptance.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2097_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2025/supp/2097_supp.zip
Link to the Code Repository
https://github.com/alanqrwang/morphldm
Link to the Dataset(s)
N/A
BibTex
@InProceedings{WanAla_Generating_MICCAI2025,
author = { Wang, Alan Q. and Huang, Fangrui and Trang, Bailey and Peng, Wei and Abbasi, Mohammad and Pohl, Kilian and Sabuncu, Mert R. and Adeli, Ehsan},
title = { { Generating Novel Brain Morphology by Deforming Learned Templates } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15961},
month = {September},
page = {205 -- 215}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes a 3D brain MRI generation method based on latent diffusion models (LDMs). The proposed model generates novel images by applying synthesized deformation fields to a learned template. Experimental results show that the models superior to generative baseline methods .
- 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 paper proposes a novel image generation method based on learned deformations from a learned atlas. this formulation is closer to previous works on computational anatomy that characterize anatomical variability using constrained spaces rather than complete randomness (ex. Kendal shape spaces).
-
I like the idea of incorporating external attribute-based priors (like age, sex) in the latent space. this is practically very useful in medical imaging and enhances model interpretability
-
the method is tested on a large dataset of 27K volumes.
-
the generative model is evaluated in terms of data generation quality and ability to capture age/sex characteristics on both synthetic and real data
-
the presentation of the method is clear despite being a relatively complex method
-
- 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.
-
are there any assumptions on the recovered deformation filed ? previous works on deformable atlases would stave to compute a diffeomorphic deformation field. does the model guarantees any such properties on the deformation field ? Maybe more constrained models like stationary vector field could be used.
-
loss Lsim - L1 loss is not necessarily robust to intensity variations. did the authors considered measures that are robust to intensity variations like MI ?
-
- 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 does not mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.
- 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?
The paper proposes a deformation based image generation method capable of incorporating external attributes like age/gender. The formulation is interesting and well motivated by previous works on computational anatomy based on template deformations.
- 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 introduces MorphLDM, a novel generative framework for 3D brain MRI synthesis based on latent diffusion models (LDMs) that departs from traditional image-space generation. Instead of synthesizing full brain images directly, MorphLDM generates deformation fields applied to a learned anatomical template, enabling the model to better capture morphological variations such as age, sex, or disease effects. Unlike prior works that fix the template or rely on precomputed deformations, MorphLDM jointly learns both the template and deformation fields, conditioned optionally on subject metadata.
- 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 paper proposes MorphLDM, a new formulation of 3D brain MRI generation that departs from typical direct image synthesis.
Instead of generating full images voxel-by-voxel, MorphLDM synthesizes deformation fields that are applied to a learned anatomical template. Unlike prior works that use fixed templates or rely on precomputed deformation fields (e.g., [3, 5, 36]), MorphLDM learns both the template and deformation fields simultaneously in a data-driven way.
- 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.
The approach assumes that each brain image can be generated by deforming a common template. However, this may not hold in pathological cases involving lesions, tumors, or signal anomalies, where intensity differences are not explained by geometry alone.
The authors mention this limitation briefly, but do not provide empirical evidence or robustness evaluation under such scenarios.
- 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
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?
The proposed MorphLDM introduces a novel approach to brain MRI generation by synthesizing deformation fields applied to a learned template, which leads to improved anatomical realism and condition-specific generation
- Reviewer confidence
Not confident (1)
- [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
The authors cast the generation of novel anatomy into a registration framework, so that instead of predicting the anatomy directly, a deformation field is predicted that can be applied to a template. The template itself is the result of a generative process that can be conditioned on anatomical attributes. A Latent Diff. Model then generates the anatomy from the template incorporating the learned realistic deformations.
- 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 paper is well written and organized.
- The authors show a broad background knowledge, and the methodological contribution appears well founded and systematically constructed.
- Validation experiments and comparison with own baseline (LDM without templates) as well as other generative models are complete and comprehensive
- Studies of performance across age groups are convincing
- The general idea of casting the generative process into a registration approach is interesting (downsides discussed below and also noted by the authors)
- 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.
- The major weakness to me is the limited applicability to anatomy that can be represented in an atlas. In the given study, only the brains (skull-stripped) of only the normal controls have been used, ensuring a high level of similarity. The authors note this limitation themselves in their footnotes, but unfortunately don’t offer insights into how this might be alleviated.
- In the study of volumes for brain regions, they note that MorphLDM is “worse” than LDM for some, without trying to explain what might be particular about these regions.
- I note strange artefacts in the Fig.2 depiction of own results. While the anatomy looks more detailed, in the sagittal plane, the anterior region looks as if there is a “bias field” brightening the region too much; also the white matter/gray matter foldings look inconsistent, as if it was not generated in 3D but 2D (which I know it isn’t). This surprises me since there is no such look in the two other planes.
- 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
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?
This is mature work advancing the field of research. It is based on solid understanding of all adjunct fields, not “simply throwing deep learning” onto some problem.
- Reviewer confidence
Somewhat confident (2)
- [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