List of Papers Browse by Subject Areas Author List
Abstract
Analyzing fetal body motion and shape is paramount in prenatal diagnostics and monitoring. Existing methods for fetal MRI analysis mainly rely on anatomical keypoints or volumetric body segmentations. Keypoints simplify body structure to facilitate motion analysis, but may ignore important details of full-body shape. Body segmentations capture complete shape information but complicate temporal analysis due to large non-local fetal movements. To address these limitations, we construct a 3D articulated statistical fetal body model based on the Skinned Multi-Person Linear Model (SMPL). Our algorithm iteratively estimates body pose in the image space and body shape in the canonical pose space. This approach improves robustness to MRI motion artifacts and intensity distortions, and reduces the impact of incomplete surface observations due to challenging fetal poses. We train our model on segmentations and keypoints derived from $19,816$ MRI volumes across $53$ subjects. Our model captures body shape and motion across time series and provides intuitive visualization. Furthermore, it enables automated anthropometric measurements traditionally difficult to obtain from segmentations and keypoints. When tested on unseen fetal body shapes, our method yields a surface alignment error of $3.2$ mm for $3$ mm MRI voxel size. To our knowledge, this represents the first 3D articulated statistical fetal body model, paving the way for enhanced fetal motion and shape analysis in prenatal diagnostics. The code is available at https://github.com/MedicalVisionGroup/fetal-smpl .
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/4748_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2025/supp/4748_supp.zip
Link to the Code Repository
https://github.com/MedicalVisionGroup/fetal-smpl
Link to the Dataset(s)
N/A
BibTex
@InProceedings{LiuYin_Fetuses_MICCAI2025,
author = { Liu, Yingcheng and Wang, Peiqi and Diaz, Sebastian and Abaci Turk, Esra and Billot, Benjamin and Grant, P. Ellen and Golland, Polina},
title = { { Fetuses Made Simple: Modeling and Tracking of Fetal Shape and Pose } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15970},
month = {September},
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper fits the infant Skinned Multi-Person Linear model (infant SMPL) to in utero MRI time series to produce a statistical model of gestating fetus shape and pose.
Essentially, the model is a mean surface point cloud, principal component displacements, articulated joint key points, and blending weights which describe how joint articulations affect shape.
An optimization procedure where pose and shape parameters are decoupled and optimized iteratively in an interleaved fashion is described. According to experimental results, this decoupling is required to obtain the best model fit.
Experiments show that the proposed fetus model fits in utero time series better than the original infant SMPL, even when the infant SMPL is fit using the decoupled optimization scheme.
- 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.
I have concerns about this paper which I will describe below, but I believe the project itself has many strengths.
First, this is an underserved research area that has tremendous potential for impact. Such a model may provide non-invasive diagnostic tools for developmental disorders at an earlier gestational age. Measurement of fetus anatomy from 2D ultrasound in the clinic is prone to error and bias, but full 3D measurements obtained computationally and referenced to a statistical model would be much more objective and potentially reveal subtleties that manual measurements might miss.
Second, the data modality here is very challenging to work with. Fetuses don’t sit still for you during imaging, even if you ask very nicely. And for time series imaging, large volumes must be imaged at a fast frame rate, which leads to lower spatial sampling rates and increased noise. Using a statistical model to infer shape and pose overcomes these challenges by encoding abundant prior knowledge. The fetus shapes and positions extracted by the model and presented in the paper look very convincing.
- 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.
This paper is a bit complex for me to review, because I believe the project is excellent, but that the paper is not well written.
The presentation of the model mathematics and the optimization procedure are both very poor. Every equation is inline, which makes following the explanation and referencing prior statements very difficult for the reader. Many variables are shown but left unexplained. The presentation of model and optimization concepts is disordered between sections, i.e., there is not a consistent logical flow of concepts section to section. I will provide examples for each of these problems.
No specific example is needed to see that every equation presented is inline. Examples of equations which are often broken out and enumerated are key statements in the model definition (several are shown inline in section 2.1) and optimization objective functions (several are shown inline in section 2.2).
A few examples of unexplained and/or confusing variables:
In section 2.1, we have \theta \in R^72. It’s not clear where this 72 comes from. If \theta are rotation vectors e.g., (\theta_x, \theta_y, \theta_z) and there is one for each of K = 23 key points, then that would be 69 values. Perhaps there is a global rotation or something as well? Or maybe the 72 or K = 23 is a typo? Or maybe something else is just not clear? But that is my point. The explanation is unclear.
In section 2.1 we also have B_p(\theta) = \sum_{k=1}^{9K} (R_k(\theta) - R_k(\theta^*))P_k. The R values in this expression are never explained anywhere in the paper. Further, P_k is described as “the vectorized rotations matrix for joint k” which, at least to me, implies that it is a vector of length 9 (vectorized 3D rotation matrix). The above equation is a linear combination of vectors of length 9, and is therefore itself a vector of length 9. But later, this whole expression is added to T_s to make T_p, which is supposedly a vector of length 3N (the entire surface point cloud). I have no idea what I’m missing here - which again is the point - the explanation is just not clear.
In the initialization section, it is stated, “… and optimize the pose parameters \theta_{i,j}, translation t_{i,j}, …” which is the first mention in the entire paper of any variable t_{i,j} or any specific translations for that matter.
I find the section on initialization to be the most opaque and to also illustrate my point about the presentation of concepts being very disordered. There is no effort to correspond the order of variable definitions in section 2.1 with the order of variable groupings for optimization presented in this initialization section; extra effort in that regard would have made these sections much more accessible. This, combined with the inline presentation of the objective functions and the run on sentences explaining each term in a disordered way is just befuddling.
Overall, the mathematics seem sound to me, but I can’t confidently say for sure that they make sense because they are not presented in a way that is accessible without having to reach into background materials to explain them on the authors behalf. Which in my opinion is asking too much of your reviewer.
I realize this is primarily a criticism of the communication, rather than the scientific merit, but I can’t confidently evaluate the scientific merit if it is poorly explained. Moreover, the authors do not release their model or any source code. So this paper has to stand for the entire project reproducibility, and I am sure that I could not reproduce this work de novo from the existing descriptions.
- Please rate the clarity and organization of this paper
Poor
- 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 provide sufficient information for 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
Again - I have a hard time determining a recommendation for this paper. I believe the project is excellent, but the communication of that project through this paper is very poor.
Regarding reproducibility, the paper does not mention anywhere that the fetal model or source code would be released. So the work is completely irreproducible.
- 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.
(3) Weak Reject — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I would love to see this project published. I believe it has excellent impact potential. I’m very interested to see it applied to clinical studies looking for biometrics in developmental disorders. I would love to see this kind of technology more available to doctors and parents.
But, since neither the model itself nor any source code are released with the paper, the paper itself would stand as the only way the project can be accessed by the community. And the paper itself just does not communicate the fundamentals of how this project was done sufficiently.
- Reviewer confidence
Very confident (4)
- [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
The paper introduces a 3D articulated statistical fetal model based on SMPL. The model is trained on MRI time-series data and enables consistent surface correspondences across frames. The paper proposes a pose-shape decoupling strategy, estimating pose and canonical shape in separate steps.to overcome the limitations posed by low-resolution, noisy fetal MRI data.
- 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.
- It claims to be the first 3D articulated statistical fetal body model, which combines both shape detail and temporal coherence in a single template model.
- The model is trained on segmentations and keypoints derived from 19, 816 MRI volumes of 53 subjects.
- Decoupling pose and shape estimation to improve modeling.
- The model achieves low alignment errors for unseen shapes and poses, and outperforms adapted infant SMPL models.
- 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.
- Poof readability especially for method section, which cause confusion. (a) 2.1 the linear skinning equation misses operator for k. (b) 2.2 initialization, the last sentence is confusing. (c) 2.2 Step2, λj is not used in any of the equations of the first paragraph. (d) 2.2 New Subjects. “We then unpose the vertices the canonical space”? The sentencing is incomplete and confusing. (e). 3. Experiments and Results, Baselines. “To We evaluate…”
- The beginning of the paper mentions that the model is trained with 53 subjects. However, the experiments section brings up another dataset of 11 subjects. It is unclear how the clinical dataset is used.
- There is limited novelty with the proposed method, which is based on the infant SMPL model. It is an incremental adaptation rather than a innovative development.
- Data limitation for pathological cases.
- The concept of a shared canonical pose is adapted from SMPL-based modeling, but fetal poses in MRI are far more variable and contorted compared to adult or infant scans.
- Please rate the clarity and organization of this paper
Poor
- 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 provide sufficient information for 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.
(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?
-
Limited novelty, but valuable application While the method itself largely builds on existing techniques (e.g., template-based modeling and skinning), its application to fetal shape analysis is relatively novel. As a pilot study, it provides a useful starting point for future research.
-
Poor readability The paper suffers from significant clarity issues, particularly in the methods section, which raises concerns about the authors’ grasp of the underlying equations.. To improve clarity, the paper would benefit from a structured presentation of the methodology, such as a pseudocode block or a step-by-step algorithm table.
In summary, it is potentially useful idea for fetal shape modeling, but poorly communicated and weakly validated. Needs more clarity, rigor, and justification to make its contribution convincing.
-
- Reviewer confidence
Very confident (4)
- [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
This paper introduces the first 3D articulated statistical model of the fetal body, adapted from the SMPL framework. The authors propose a novel training pipeline for fetal MRI time series that decouples shape and pose estimation through a multi-step unposing strategy. The model is trained on over 19,000 MRI volumes from 53 subjects and enables robust full-body shape tracking, anthropometric measurements, and temporal motion analysis in challenging fetal MRI data.
- 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 presents a compelling and well-executed methodological contribution by adapting the SMPL framework for fetal MRI, an application domain with high clinical relevance and technical challenges. The key innovation lies in the robust unposing of segmentation data across time to separate shape and pose estimation, enabling accurate modeling despite motion artifacts, geometric distortions, and occlusions common in fetal MRI.
The training is performed on a large dataset (19k volumes), which supports strong generalization and statistical robustness. The authors validate their approach both quantitatively (via Chamfer distance) and clinically (via correlation of anthropometric measurements with gestational age), and demonstrate that their model substantially outperforms downscaled infant body models.
The authors also emphasize interpretability, providing clear visualization of principal components and clinical shape statistics. The methodological pipeline is well explained, reproducible, and shows thoughtful adaptation of prior SMPL-related work to a difficult medical domain.
- 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.
There are no major methodological flaws. However, one limitation is the relatively small number of test subjects in the clinical evaluation cohort. While this does not compromise the novelty or rigor of the method, broader clinical validation would strengthen the translational impact.
In addition, although a baseline comparison is made with infant SMPL (with and without unposing), inclusion of more recent shape modeling baselines or alternative non-parametric surface tracking approaches (e.g., deformation-based registration or neural fields) could offer a broader context for evaluation.
- 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
nice work!
- 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.
(6) Strong Accept — must be accepted due to excellence
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This paper introduces the first articulated statistical body model for fetal MRI, addressing a clear unmet need with high clinical relevance. The method is technically sound, well-validated on a large dataset, and significantly outperforms existing alternatives. Its applications to motion tracking and fetal growth assessment are impactful. The paper is well-written, innovative, and highly relevant to the MICCAI community, exactly what I want to see in the conference.
- Reviewer confidence
Very confident (4)
- [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
We sincerely thank all reviewers for their thoughtful and constructive feedback. We are encouraged that our work was recognized as “a valuable application” (R1), with “excellent impact potential” (R3), and as “addressing a clear unmet need with high clinical relevance” (R4). In the camera-ready version, we will improve clarity by restructuring the method section, clarifying equations and variable definitions, and correcting typographical errors (R1 and R3). To support reproducibility, we will release the code and model subject to licensing and model use agreements (R1). We are grateful for the reviewers’ time and effort to provide feedbacks. And we are looking forward to sharing this work with the community.
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