List of Papers Browse by Subject Areas Author List
Abstract
After birth, the cranium and facial skeleton undergo rapid growth. Routine postnatal assessment is crucial for the early identification of craniofacial deformities, often characterized by asymmetric growth patterns. However, a comprehensive 3D shape model capturing both soft tissue and bony structures during early craniofacial development does not yet exist. We introduce the first integrated 3D shape model of the infant head and skull, constructed from a large dataset of photogrammetric scans complemented by a smaller set of computed tomography scans. Our INfant CRANial (INCRAN) model captures detailed facial expressions and overall cranial shape variations, incorporating the most advanced representation of cranial sutures on the underlying skull to date. By mapping cranial measurements to the model’s latent space, we further distinguish various craniofacial deformities from normal shape variations, enabling automated diagnosis and correction proposals. Additionally, we propose a novel method for constructing a multi-linear model from an uncontrolled expression space by projecting an autoencoder back into PCA space, thus enhancing model interpretability. INCRAN supports growth monitoring and holds potential for improving infant healthcare and craniofacial treatment strategies.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2543_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2025/supp/2543_supp.zip
Link to the Code Repository
https://github.com/tillns/INFACE
Link to the Dataset(s)
N/A
BibTex
@InProceedings{SchTil_MultiLinear_MICCAI2025,
author = { Schnabel, Till N. and Lill, Yoriko and Benitez, Benito K. and Krief, Gaspard and Tapia Corón, Sebastián and Prüfer, Friederike and Metzler, Philipp and Mueller, Andreas A. and Gross, Markus and Solenthaler, Barbara},
title = { { Multi-Linear 3D Craniofacial Infant Shape Model } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15969},
month = {September},
page = {337 -- 347}
}
Reviews
Review #1
- Please describe the contribution of the paper
- The authors propose a method for constructing a shape model of the infant skull, mandible, and soft-tissue cranium, aiming to capture their morphological features and variations.
- They investigate the significant correlations among these domains, including the fusion of key sutures, to understand developmental relationships.
- The study employs a multi-linear PCA model to disentangle factors such as identity, expression, and age-related changes, providing a comprehensive analysis of shape variability.
- 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 manuscript is well written, presenting the methodology clearly and coherently.
- The proposed approach is straightforward and grounded in PCA statistics, utilizing linear methods that enhance interpretability and avoid the opacity often associated with machine learning black-box 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.
- The literature review is not comprehensive, omitting some critical studies on skull shape modeling and 3D photogrammetry, such as recent works on craniosynostosis and developmental shape analysis.
- The results could be better organized to effectively demonstrate the model’s capabilities; some figures may benefit from restructuring for clarity.
- The reliance on manual landmark annotations for alignment poses scalability challenges, as manual processes are time-consuming and prone to variability.
- Please rate the clarity and organization of this paper
Satisfactory
- 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
- The purpose of studying baby face expressions is unclear; it would be beneficial to clarify whether there is a clinical motivation for including this attribute in the analysis.
- Visualization in Figure 3 could be improved by displaying axial, sagittal, and coronal views to better illustrate shape differences.
- The age prediction analysis lacks comparison with simpler baseline models, such as predicting the mean age, and does not report metrics like mean absolute error, making it difficult to assess the practical utility of the approach. The clinical significance of errors around ±4.7 months should be discussed, especially considering pediatric developmental variability. Further analysis could explore which features are most predictive and whether additional anatomical markers could improve predictions.
- The dependence on manual landmark annotations for alignment is a critical limitation which needs to be discussed. Have the authors considered some ablation studies assessing the impact of misalignments on PCA based statistical performance?
- While the authors claim that this is the first application of PCA attribute correlation for diagnosing and correcting deformed infant craniums, this statement may be misleading, as prior studies have used PCA-based z-scores for classification and diagnosis of cranial deformities. Clarification and contextualization of this claim are recommended.
Some references authors should consider (and there are more):
- Bhalodia, Riddhish, et al. “Deepssm: A blueprint for image-to-shape deep learning models.” Medical image analysis 91 (2024): 103034.
- Beiriger, Justin W., et al. “Craniorate: An image-based, deep-phenotyping analysis toolset and online clinician interface for metopic craniosynostosis.” Plastic and Reconstructive Surgery 153.1 (2023): 112e-119e.
- Schaufelberger, Matthias, et al. “A radiation-free classification pipeline for craniosynostosis using statistical shape modeling.” Diagnostics 12.7 (2022): 1516.
- Elkhill, Connor, et al. “Geometric learning and statistical modeling for surgical outcomes evaluation in craniosynostosis using 3D photogrammetry.” Computer methods and programs in biomedicine 240 (2023): 107689
- 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 recommend a weak reject because, while the manuscript presents a promising and well-explained methodology, it lacks a comprehensive review of related literature and relies on manual processes that limit scalability. Additionally, the results and visualizations could be better structured to more clearly demonstrate the model’s capabilities and clinical relevance.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
After carefully reviewing the authors’ rebuttal, I am updating my score. The authors have addressed my concerns by committing to expand the literature review, clarify their claims of novelty, and release code and trained models to improve reproducibility. They have also provided justification for their manual annotation process and will add baseline comparisons and clinical context to the age prediction analysis.
Review #2
- Please describe the contribution of the paper
This study proposes an integrated 3D shape model of the infant head and skull. The model is generated from a large dataset of photogrammetric scans and is further complemented by a smaller set of CT scans, offering a comprehensive representation of infant craniofacial morphology.
- 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.
Clinical Significance The developed 3D shape model of the infant head holds substantial clinical value by potentially enabling automated diagnosis and treatment planning for various craniofacial deformities.
Enhanced Infant Morphable Model The authors have adopted and further improved the previous infant morphable model to capture not only identity and expression but also time-related factors, which is crucial for modeling developmental changes.
Multilinear PCA-like Model The final integrated model maintains the advantages of traditional PCA, including orthogonality, variance sorting, and interpretability, while extending its application to infant craniofacial data.
- 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.
Condensed Methodological Descriptions Due to space constraints, many of the adopted methods are referenced rather than fully explained. In addition, the condensed descriptions of the methodology can lead to confusion and hinder a full understanding of the techniques applied.
- 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
Landmark Annotation on 3D Data The study reports that 118 landmarks were annotated for 3D photo scans, and 360 landmarks for skull meshes derived from CT scans. Were all these landmarks digitized manually? How were these landmarks defined? Given the high number of points, it might be more appropriate to refer to them as feature points rather than traditional anatomical landmarks. Visualizing these points on a representative surface would greatly aid readers in understanding their distribution and definition.
Definition of the Skull Template Please provide a clear definition of the “skull template” used in the study. Is the template generated as an average model of the dataset, or is it derived from a randomly chosen subject? Clarification on this aspect would enhance the reproducibility of the work.
Joint Alignment via Generalized Procrustes Analysis The manuscript states that “after registration, head and skull mesh pairs from the CT dataset Chinit,Csinit ,Csinit undergo joint alignment via generalized Procrustes analysis on the soft-tissue scalp.” It appears that this step is intended to align all subjects into a common coordinate space. Please clarify this process in the manuscript, including which subjects or templates were used as a reference for alignment.
Correlation Between Cranial Measurements and Disentangled Model Factors The study would benefit from a more detailed explanation of how the correlations between key cranial measurements and the factors of the disentangled model were calculated.
- 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 integrated 3D shape model of the infant head and skull is a promising contribution with clear clinical implications, particularly in the realm of automated diagnosis and treatment planning for craniofacial deformities.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
The authors addressed the reviewers’ questions and committed to improving the manuscript accordingly.
Review #3
- Please describe the contribution of the paper
The authors propose a novel methodology to conceive a 3D shape model that accounts for infants’ sutures development and better anatomy representation, integrating not only the skull, but soft-tissue and mandible. Moreover, the study integrates more parameters age-related, as well as identity attributes, making the whole study more comprehensive and generalized.
- 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.
1) This study incorporates better anatomy of the infants’ cranium, not only the skull, but the soft-tissue and also the mandible; it also comprises both CT and photogrammetric datasets, providing a richer representation (for example, Schaurich et al. https://doi.org/10.1038/s41598-024-77315-8 and O’Sullivan et al. https://doi.org/10.1016/j.bonr.2021.101154 both propose 3D shape models encompassing CT scans from the infants’ skulls); 2) The methodology is extensively detailed and carefully explained, especially the advantages of using linear autoencoders for compatibility with the linear regressors; 3) The authors also elaborate on the potential clinical impact of their proposed technology especially in terms of diagnosis and potential aid in treatment planning; 4) The paper is exceptionally well written and coherent and presents high quality images that help the reader further understand what is explained;
- 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.
1) The authors do not address the outputs of the study in the abstract, which leads to a lack of clarity; the results should be referred to so that the reader can understand the impact of the proposed methodology; 2) The authors do not disclose whether the landmarking was made manually or in an automated manner; 3) Throughout the paper, it is not entirely clear how the proposed method is better than previous works; the reader understands that the application is more comprehensive, because of different anatomical elements, craniofacial variations and latent spaces, however, there is not an extensive comparison to previous works that allows understand on the novelty of this method; 4) The training details are scattered throughout the paper; I would recommend a brief section dedicated to the description of the training parameters, for purposes of reproducibility; 5) The evaluation metrics could be disclosed in the section ‘3. Results and Applications’ on a brief sentence.
- 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
–> On page 1, in section ‘Introduction’, line 6, there are two periods before and after the referencing (in ‘…management.[20].’); –> On page 5, in section ‘2.4 Linear Disentanglement’ the authors reference previous work in two manners: by merely reference (‘In [21],…’ and ‘proposed by [26].’) and by stating the name of the authors (‘… Kunin et al. [10]…’); I would recommend following a unique manner of referencing previous work for better presentation; –> On page 10, in section ‘References’, the DOI link on reference 21 leads to a ‘DOI not found’ page;
- 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 study presents a promising application in the medical domain. If tested further, it could be of great value for treatment of infants. The methodology integrates better anatomy, head-skull correlations and disentangled spaces of individual-related attributes - expression, age and identity. It is the most comprehensive approach in this area of research.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
Considering the rebuttal and that one reviewer made also a clear accept (the other made a weak reject), I stay with my decision accept.
Author Feedback
Review of Literature (MR, R1) We thank the reviewers for their feedback and suggested references, which we will discuss in the revision. The three craniosynostosis-focused studies add valuable context to one of our application areas (Sec. 3.3). While these model CT (Beiriger’23) or surface data (Schaufelberger’22; Elkhill’23), our method unifies multiple anatomical modalities and disentangles age, identity, and expression, supporting broader applications beyond classification or metric extraction. We will cite recent work on automated landmarking and segmentation (Torres’19, Bhalodia’24) in our discussion of scalability. O’Sullivan’21 and Li’15 are already cited ([24], [13]).
Attribute Correlation and Novelty (R1) We will rephrase our claim on PCA attribute correlation to better align with related work. Our method differs in the parameters being regressed: by regressing common cranial measurements—rather than binary malformation labels as in prior studies—we aim to define a more continuous direction in shape space. More broadly, our contributions lie in integrating skull, mandible, and surface into a unified statistical model with a disentangled latent space for age, identity, and expression. Unlike prior models that are modality-specific or static, ours supports cross-modality inference, temporal shape progression, and metric extraction across both soft tissue and bone, enabling applications beyond the scope of existing approaches.
Comparisons (MR, R1, R3) Direct comparisons are limited by the lack of publicly available 3D morphable head models and standardized datasets. Schaufelberger’22 offer the only accessible model we know of, focused on surface-only craniosynostosis data. While their model is available, differences in the underlying (non-shared) patient data make direct comparison challenging. Moreover, we consider our joint head–skull model, designed for broader applications, complementary to prior work.
Manual Landmarking (R1, R2, R3) Given the variability in scan quality across our datasets, we opted for manual annotation, as many automated methods were not reliable in this context. Although we did not perform an ablation on alignment sensitivity, consistent manual alignment proved important for robust modeling and can serve as a baseline for future automation efforts. “Feature points” better describes our annotations, as many lie along curves or ridges rather than anatomical landmarks. While the point count is high, most were efficiently placed along structured paths. This is shown in the supplementary video, and we will add an image in the revision.
Reproducibility (R1, R2, R3) Due to privacy protections for infant patient data, we cannot release the raw dataset. However, we will share the trained model and code required to reproduce key results. We will also revise the paper to clarify the methodology and better support reproducibility.
Age Prediction (R1) We can add a simple baseline (mean-age prediction) and report mean absolute error to improve interpretability. The observed ±4.7-month error falls within typical developmental variability, but we will include a brief discussion on its clinical relevance. We agree that analyzing latent space contributions is worthwhile and consider this a promising direction for future work.
Expressions (R1) Expression disentanglement has been explored in prior work [21], where it was shown to be useful for neutralizing facial expressions to enable more consistent comparisons across scans (e.g., pre- vs. post-operative) and improving the reliability of standardized measurements across subjects. Our inclusion of this attribute supports these potential applications.
Skull Template (R2) The skull template is artist-designed from a representative patient case. Closed sutures are linked by virtual edges with high stiffness between initially co-located vertices and disconnected components, and dense landmarks ensure accurate deformation to input cases.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- 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”.
Being an expert on 3D face correspondence and shape models, I have concerns about the novelty of the contributions of this paper. I share the concerns flagged by R1, that some important papers have not been cited. For example:
O’Sullivan, Eimear, et al. “The 3D skull 0–4 years: a validated, generative, statistical shape model.” Bone reports 15 (2021): 101154.
Li, Zhigang, et al. “A statistical skull geometry model for children 0-3 years old.” PloS one 10.5 (2015): e0127322.
Torres, Helena R., et al. “Deep learning-based detection of anthropometric landmarks in 3D infants head models.” Medical Imaging 2019: Computer-Aided Diagnosis. Vol. 10950. SPIE, 2019.
Furthermore, lack of comparison with prior art makes it difficult to assess the significance of the results.
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
Well-written paper with limited methodological novelty but with an interesting application scenario that’s underrepresented at MICCAI. All reviewers are in favor of the paper after the rebuttal phase.