Abstract

Predicting individualized perinatal brain development is crucial for understanding personalized neurodevelopmental trajectories, however, remains challenging due to limited longitudinal data. While population based atlases model generic trends, they fail to capture subject-specific growth patterns. In this work, we propose a novel approach leveraging Implicit Neural Representations (INRs) to predict individualized brain growth over multiple weeks. Our method learns from a limited dataset of less than 100 paired fetal and neonatal subjects, sampled from the developing Human Connectome Project. The trained model demonstrates accurate personalized future and past trajectory predictions from a single calibration scan. By incorporating conditional external factors such as birth age or birth weight, our model further allows the simulation of neurodevelopment under varying conditions. We evaluate our method against established perinatal brain atlases, demonstrating higher prediction accuracy and fidelity up to 20 weeks. Finally, we explore the method’s ability to reveal subject-specific cortical folding patterns under varying factors like birth weight, further advocating its potential for personalized neurodevelopmental analysis.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3652_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{DanMai_Predicting_MICCAI2025,
        author = { Dannecker, Maik and Rueckert, Daniel},
        title = { { Predicting Longitudinal Brain Development via Implicit Neural Representations } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15970},
        month = {September},
        page = {481 -- 491}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The primary contribution of the paper is the novel method leveraging implicit neural networks they proposed to predict individualized growth patterns and trajectories for neonatal and fetal brains. The key difference from previous works in the field is that previous works utilized atlas based methods to calculate average trajectories. The proposed model can work on both fetal and neonatal brains, and can predict up to 20 weeks into both future and the past, based on subject specific morphologies.

  • 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 believe the key strengths of the paper lies in its novel methodology. In contrast to previous methods, the proposed model makes individualized predictions specific to each subject. The model also takes into account conditional external factors such as age and weight at birth. The proposed model is also robust and adapts well to cross-domain prediction, given that it can handle both fetal and neonatal brains and both future and past predictions. The fact that the model does not require a resampling of the scans to be in the same resolution also is a strength that would help in real world application.

  • 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.

    Currently, the model has not been tested on unseen subjects, as the authors do mention in the paper. This is a limiting factor for clinical translation; it is difficult to understand the generalizability of the proposed. The need for high segmentation quality, which I believe is more challenging to acquire with fetal and neonatal brain scans in particular, also make real world application difficult as is. Overall, I think some more work to convince the readers of the interpretability of the model would benefit clinical translation. In conclusion, while I think the work is promising, there are missing components that are needed before it becomes feasible for a possible clinical application.

  • 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.

    (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?

    I think the work is promising, but a little more analysis would benefit it. The authors acknowledge the limitation in the generalizability until they test the model on unseen data. Given this limitation, such as an ablation study to see which component of the model contributes the most to the performance. Or, an uncertainty estimation to better improve clinical application or an analysis to confirm interpretability of the model would at least partially overcome the limitation of the work as is without external unseen data and relatively small dataset size.

  • 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



Review #2

  • Please describe the contribution of the paper

    The authors presented a spatiotemporal atlas designed to predict individualized growth patterns of fetal and neonatal brains. Their method enables subject-specific forecasting of brain development up to 20 weeks into the future or past for both fetal and neonatal imaging data, allowing for the modeling of developmental trajectories that begin in the fetal stage. By integrating conditioning on external factors—such as birth age or birth weight—the model can simulate brain development under customized conditions. Despite being trained on a relatively small dataset of 100 subjects, each with only one follow-up session, the model achieves accurate individualized brain growth predictions using a single calibration scan. It outperforms traditional population-based atlases in predicting subject-specific neurodevelopment.

  • 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) The use of implicit neural representations to generate subject-specific atlases represents a novel contribution in the field. 2) The experimental tasks were well-designed to span both fetal and neonatal stages, enabling prediction of brain development in both past and future directions. 3) The image analysis pipeline was clearly described, with step-by-step details and inclusion of quantitative evaluation methods. 4) The results demonstrate that the proposed model outperforms conventional atlas-based approaches in predicting individualized brain development.

  • 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 results of Fig 3 and 4 are not clear. Even though subject-specific images for multiple time points are generated and reflect brain development patters (volume increase and cortical thinning), the ground truth dataset to validate these all images were not available (as I understand).

  • 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 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.

    (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 paper presents several strengths, despite a couple of results that remain unclear.

  • 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

    This paper introduces a novel framework leveraging Implicit Neural Representations (INRs) to predict individualized longitudinal trajectories of fetal-neonatal brain development. Key contributions include:

    1. Subject-Specific Prediction: The first method to model individualized brain growth trajectories across prenatal and postnatal periods (up to 20 weeks), addressing the challenge of capturing rapid and heterogeneous morphological changes during perinatal development.
    2. Condition-Aware Simulation: A pioneering approach to simulate neurodevelopmental trajectories under varying external factors (e.g., birth age, birth weight), enabling the study of their influence on cortical folding and brain morphology—an underexplored capability in prior works.
    3. Cross-Domain Bridging: The first framework to predict brain development trajectories that seamlessly transition between fetal and neonatal domains, overcoming the anatomical discontinuity introduced by birth.
  • 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. The accuracy of the model outperforms established population-based atlases (BD-Atlas, Deepali) in metrics like PSNR, SSIM, and Dice scores, demonstrating superior fidelity to individual anatomy.
    2. Due to the nature of INS, the model achieves accurate predictions using a limited dataset (70 training subjects), making it practical for scenarios with scarce longitudinal data.
    3. The model incorporates conditional tokens to model external factors, allowing simulations of “what-if” scenarios (e.g., varying birth weight).
    4. The model requires only a single calibration scan for predictions, aligning with clinical constraints where repeated imaging is rare.
  • 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. Limited Generalization: Predictions are restricted to subjects included in training; the method cannot yet generalize to entirely unseen subjects without test-time optimization.
    2. The training process on only 70 subjects raises concerns about robustness and scalability, particularly for rare or heterogeneous developmental conditions.
    3. Fetal MRI protocols (e.g., slice thickness, super-resolution methods) differ significantly from neonatal imaging, yet their harmonization and potential biases are not rigorously analyzed—a critical gap given the influence of image quality on INR performance.
  • 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.

    (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?
    • Technical Innovation: The integration of INRs for longitudinal brain modeling and conditional simulations represents a significant methodological leap.
    • Empirical Validation: Strong quantitative results and cross-domain predictions substantiate the framework’s superiority.
    • Clinical Potential: Single-scan calibration and conditional simulations address unmet needs in personalized neurodevelopmental analysis.
    • Limitations: Restricted generalizability and cohort size temper enthusiasm, highlighting the need for future work on scalability and protocol standardization.
  • 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 rebuttal was effective, and I have no additional questions.




Author Feedback

We thank the Area Chair (AC) and Reviewers (R1, R2, R3) for their valuable feedback. We are pleased and encouraged that the merit and relevance of our work has been recognized. Below, we consolidate and address the remaining concerns.

  1. Novelty (AC) We acknowledge similarities to prior work [4]. Yet, our work innovates in several aspects:
    • Learnable Tokens: Automatically learn optimal, multi-dimensional representations without predefined anatomical parameters, enabling richer conditioning.
    • Contrastive Constraint: Orthogonalized tokens ensure disentanglement between correlated variables (birth-weight vs birth-age)
    • Unification Across Domains: Shared latent codes across fetal-neonatal sessions unify prenatal and postnatal domains bridging the event of birth.
    • Individualized and Conditionable Predictions: Our work targets individualized predictions. Combined with conditioning on external factors (e.g. birth-weight, birth-age) via learnable tokens, the presented work enables simulations of “what-if” scenarios (as R1 neatly put it) for individuals, addressing a critical and underexplored gap. We will clearly rework the contributions section to highlight these points.
  2. Generalization, Robustness to Domain Shifts (R1, R2, AC) We acknowledge concerns regarding limited generalization to unseen subjects and domain shifts in data. Future work will explore test-time optimization of unseen subjects. Combined with data augmentation, we are optimistic to achieve fast test-time optimization, robust to domain shifts, e.g. from MRI protocols.

  3. Interpretability and Clinical Translation (R2) We agree that interpretability is indeed crucial for clinical application. We are therefore currently exploring uncertainty quantification via a variational latent code design, enabling probabilistic predictions, i.e. a distribution over likely trajectories, enhancing transparency and enabling abnormality detection. We will integrate the findings within the scope of an extension. R2 is correct that perinatal brain MRI segmentation is challenging. Currently we use dHCP’s pipeline. However, recent segmentation methods[25] yield robust perinatal brain segmentation, even for multi-centric data, supporting clinical applicability in future.

  4. Performance, Ground Truth (GT), Birth Weight Conditioning (AC, R3) Quantitative Perf: To address AC’s concern of unsubstantial improvements: A paired t-test confirms significant improvements (p«0.01), except deepali_method1_VOL; test will be added to the manuscript. Qualitative Perf: While Fig. 3 may resemble a generic prediction (attributed to this subject’s typical anatomy), Fig. 2 clearly demonstrates capturing of individual anatomy, including brain, CSF, and ventricle morphology, even for predictions up to 19 weeks. Limited GT in Fig. 4: GT is limited to one condition+time-point per subject. Yet, observed patterns of decreasing brain size and cortical maturation with decreasing birth-weight z-scores, align closely with literature[6,29,30]. We add existing GT to Fig. 4, and further discuss related literature.

  5. Baseline Methods (AC) We recognize stronger baselines could strengthen our conclusions: However, the suggested baseline [28] while promising has critical limitations as mentioned in our related works. Limitations of [28] include:
    • Evaluation is mainly limited to predicting a subject’s previous and next week only
    • Operates in single domain (fetal)
    • No external conditioning possible
    • No validation on GT data reported As the code is not public (we contacted the authors but no response so far), together with above limitations, we chose more practical and transparent baselines within feasible research constraints.

[29] Dubois, J et al. Primary cortical folding in the human newborn: an early marker of later functional development. Brain 2008 [30] Kersbergen, K.J. et al. Relation between clinical risk factors, early cortical changes, and neurodevelopmental outcome in preterm infants. Neuroimage 2016




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”.

    The following concerns should be addressed in the rebuttal:

    1. Lack of methodological novelty. While the use of implicit neural representations to generate subject-specific images represents a novel contribution in the field, the approach is highly similar to a method proposed in a MICCAI 2024 paper [1]. Could the authors clarify what innovations have been introduced in this work, specifically in terms of generating subject-specific images, compared to the previous MICCAI method?
    2. Overly simplistic baselines. The comparison is limited to atlas images and registered known images, without including any state-of-the-art longitudinal prediction methods such as [2]. How can the superiority of the proposed method be demonstrated without stronger baselines?
    3. Limited performance. Although the proposed method achieves the best quantitative results in Table 1, the improvements are not substantial. The results in Figure 3 indicate that the INR captures a continuous growth trajectory, but there remains a noticeable gap compared to the ground truth. It is possible that the model has primarily learned a population-level growth trajectory, making it difficult to assess whether it has truly achieved subject-specific prediction. Furthermore, the results related to birth weight conditioning are insufficiently reported. Based on Figure 4, it appears that at the same gestational age, a larger birth weight corresponds to more cortical folding. Is this interpretation correct? Is there any clinical evidence supporting this observation?

    [1] Dannecker, Maik, et al. “CINA: Conditional Implicit Neural Atlas for Spatio-Temporal Representation of Fetal Brains.” MICCAI 2024. [2] Zhang, Kai, et al. “Development-driven diffusion model for longitudinal prediction of fetal brain mri with unpaired data.” IEEE Transactions on Medical Imaging (2024).

  • 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’

    N/A



back to top