List of Papers Browse by Subject Areas Author List
Abstract
Predicting long-term bone remodeling after mandibular reconstruction would be of great clinical benefit, yet standard generative models struggle to maintain trajectory-level consistency and anatomical fidelity over long horizons, particularly in low-data regimes. We introduce OsteoFlow, a flow-based framework predicting Year-1 post-operative CT scans from Day-5 scans. Our core contribution is Lyapunov guided trajectory distillation: Unlike one-step distillation, our method distills a continuous trajectory over transport time from a registration derived stationary velocity field teacher. Combined with a resection aware image loss, this enforces geometric correspondence without sacrificing generative capacity. Evaluated on 344 paired regions of interest, OsteoFlow significantly outperforms state-of-the-art baselines, reducing mean absolute error in the surgical resection zone by ~20%. This highlights the promise of trajectory distillation for long-term prediction in low-data clinical settings. Code is available on GitHub: https://github.com/hamidreza-aftabi/OsteoFlow.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/5704_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/hamidreza-aftabi/OsteoFlow
Link to the Dataset(s)
N/A
BibTex
@InProceedings{AftHam_OsteoFlow_MICCAI2026,
author = { Aftabi, Hamidreza AND Yu, Faye AND Switzer, Brooke AND Fishman, Zachary AND Prisman, Eitan AND Hodgson, Antony AND Whyne, Cari M. AND Fels, Sidney AND Hardisty, Michael},
title = { { OsteoFlow: Lyapunov-Guided Flow Distillation for Predicting Bone Remodeling after Mandibular Reconstruction } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16889},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes a flow-based generative model (OsteoFlow) for predicting long-term bone remodeling from early post-operative CT scans. The key contribution is a Lyapunov-guided trajectory distillation framework, where a student velocity field is trained to follow a registration-derived teacher trajectory.
- 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 problem of long-term remodeling prediction is high-impact and underexplored. The way authors approached it combining normalizing flows and Lyapunov function to stablize the flow seems novel and effective while avoiding training difficulties in GAN and reducing high sampling cost in diffusion models
- extensive experiments validate the effectiveness of the proposed method with respect to multiple flow-based, diffusion-based and GAN-based pipelines. Ablations also adequately illustrate the contribution of each proposed component
- 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 synthesized ROIs seem small (or perhaps low-resolution), higher-resolution synthesis may be of more clinical importance
- The authors do not provide visual comparisons with baseline methods
- The two-stage teacher and student pipeline, while understandable, seems a little bit overly complicated compared to baselines, which are all single stage models. It is also possible that the performances of other comparisons can be boosted with another stage.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 overall methodology is novel and effective, as both discussed in the paper and illustrated by the experiments.
- 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
OsteoFlow distills a rectified-flow student from a SVF registration teacher via a Lyapunov reference-tracking loss, predicting Year-1 mandibular CT from Day-5 post-op CT. It reports ~20% mid-slab bone MAE improvement.
- 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.
Clinically meaningful task, with strong motivatioms - predicting graft-host bone remodeling after mandibular reconstruction addresses a real unmet need (nonunion rates up to 37%).
Thorough ablation Table 2 with T1-vs-T2 cleanly isolating the Lyapunov formulation, and a backbone ablation (Table 3) controlling for architecture.
- 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.
Table 1 claims Wilcoxon $p < 0.05$ across runs but with $n = 3$ runs the minimum two-sided signed-rank $p$ is 0.25 suggesting that the test must be per-ROI, but patient-level clustering (344 ROIs from 120 patients) is unaddressed.
Headline ~20% bone MAE gain is mid-slab only (12 slices coinciding with the W training loss region). Full-volume bone MAE improves significantly less, abstract should name both figures as the current reporting is self-favoring.
$\alpha$ is tested only at the binary level, a finer sweep is needed to show $\alpha = 1$ is optimal, not just that any positive $\alpha$ helps.
No clinical outcome evaluation despite motivating the work with nonunion prediction. All metrics are image similarity. A model accurate on the dominant union class but failing on nonunion cases would still win these averages. Perhaps per-outcome-class / severity bone MAE.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper addresses a clinically meaningful task with a novel approach, and provides strong evidence for their Lyapunov formulation. My concerns about the statistics and outcome metrics lowers my rating.
- 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 #3
- Please describe the contribution of the paper
This paper introduces OsteoFlow, a flow-based framework for predicting Year-1 post-operative CT scans from Day-5 scans. The method distills a continuous trajectory over transport time from a registration-derived stationary velocity field teacher. Experimental results show that OsteoFlow outperforms state-of-the-art baselines.
- 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 paper builds on flow-based modeling and is specifically designed for longitudinal remodeling prediction using teacher guidance.
2.The experimental design is well-structured and demonstrates strong 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.
1.The paper lacks a figure illustrating the overall workflow of the proposed method, which would help improve clarity.
2.The results and discussion should provide more insight into the clinical impact and relevance of the proposed approach.
- 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.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- 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/2026/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 — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The experimental design is well-structured and demonstrates strong performance.
- 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
Author Feedback
We thank the reviewers for their constructive feedback. We are encouraged by the early acceptance and recognition of novelty/effectiveness (R2,R4,R5), clinical relevance (R2,R4), and strong validation/ablations (R2,R4,R5). Below, we clarify the main points raised.
Resolution/ROI (R2): We thank R2 for this observation. The data are retrospective clinical CT scans acquired during routine care, not under a controlled research protocol; resolution varied, so ROIs were resampled to 0.5 mm isotropic spacing. The ROI was localized around the resection/graft-host interface, where callus bridging, remodeling, and union/nonunion changes are expected. This reflects the limited-data setting and tests whether remodeling can be estimated from the local region alone. We clarified that HU windowing may make ROIs appear brighter while suppressing metal outliers.
Baseline Visuals (R2): We agree that qualitative baseline comparisons would be useful. Due to space limits, Fig. 2 focused on OsteoFlow predictions across union, partial-union, and nonunion patterns, while Table 1 reported quantitative baseline comparisons. Since the appendix is limited to videos, we will include baseline examples in the GitHub repository near the resection/graft-host interface.
Complexity (R2): We acknowledge the added training-time structure. This was motivated by limited data, where the SVF teacher provides a stable anatomical prior. The teacher is trained once, frozen during student training, and discarded at inference, so prediction uses only the student. We clarified that this targets low-data learning rather than inference complexity.
Statistics (R4): We thank R4 for raising this point. We clarified in the caption of Table 1 that significance was assessed using paired ROI-level test metrics. We also noted potential within-patient dependence from multiple ROIs per patient and identified patient-level or clustered analysis as a future refinement.
Reporting (R4): We thank R4 for this comment. The mid-slab result was emphasized because it corresponds to the resection/graft-host interface, where clinically meaningful remodeling and union/nonunion changes are expected. These localized changes can be diluted over the full ROI. To avoid ambiguity, we kept both evaluations and clarified the abstract/results to report both ∼19.8% mid-slab and ∼7.5% full-volume bone MAE reduction relative to cDDPM(∆).
α Sensitivity (R4): We agree this is a useful direction. We had already acknowledged this in the submission, noting that α=1 was selected empirically and detailed α sensitivity is reserved for future work due to space limits. Our aim was not to claim theoretical optimality of α=1, but to show the benefit of Lyapunov-guided correction. The ablations bracket this effect: RF-only removes Lyapunov/teacher guidance, while the privileged teacher provides the opposite hard-guidance reference.
Clinical Impact (R4,R5): We thank R4 and R5 for emphasizing clinical interpretation. Our goal is image-based remodeling prediction, which is more nuanced than binary union/nonunion classification. Although the dataset is union-dominant, Fig. 2 includes union, partial-union, and nonunion cases showing relevant patterns. We used image and bone-focused metrics to evaluate continuous change at the graft-host interface, but agree clinical outcome analysis is important. We clarified that the model is not intended for direct clinical decision-making and identified per-outcome/severity evaluation as future work.
Workflow (R5): We thank R5 for the suggestion. Fig. 1 provided a conceptual overview of preprocessing, ROI extraction, SVF teacher guidance, and Lyapunov-guided distillation. Since our novelty is mainly in the training/distillation method rather than a new architecture, Algorithm 1 presented the workflow step by step. We clarified Fig.1/caption to make the end-to-end process explicit, including Day-5 CT input, teacher-guided training, Year-1 prediction, and teacher removal at inference.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
Reviewers all agree on the overall work’s relevance, however authors should provide more interpretation of the results and discussion insights into the clinical impact and relevance of the proposed approach.