List of Papers Browse by Subject Areas Author List
Abstract
Accurate 3D dental reconstruction from sparse, uncalibrated intra-oral photographs is critical for radiation-free digital dentistry. However, traditional parametric methods lack strong anatomical priors, are restricted to limited parameter spaces, and exhibit high inference latency. Meanwhile recent generative approaches often lack geometric rigor and cross-view consistency. We propose ARCH-Net, an Aligned 3D teeth Reconstruction approach via a Counterfactual-enhanced Hierarchical framework. ARCH-Net systematically decomposes the task into three progressive stages: (1) Anatomically-constrained latent space construction to capture robust geometric priors within a learned embedding space; (2) Counterfactual-enhanced 2D-3D alignment, where a latent diffusion model is supervised by a novel counterfactual learning strategy. By contrasting predicted latent codes against mismatched negative samples, this strategy significantly amplifies the model’s discriminative power to enforce rigorous geometric consistency across sparse views; and (3) Prior-driven mesh refinement via sequential rigid and non-rigid registration to ensure the topological integrity of final high-resolution models. Experimental results demonstrate that ARCH-Net significantly outperforms state-of-the-art methods in both reconstruction accuracy and computational efficiency. Furthermore, our approach is robust to varying imaging conditions, offering a reliable, radiation-free solution for precision orthodontic planning and real-time digital diagnostics.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/0794_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{DenQin_ARCHNet_MICCAI2026,
author = { Deng, Qingxin AND Jiang, Landu AND Yan, Yuzheng AND Luo, Wuman AND Zhang, Dian},
title = { { ARCH-Net: Aligned 3D Teeth Reconstruction via Counterfactual-Enhanced Hierarchical Framework } },
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
The paper tackles an important clinical problem and presents a plausible hierarchical method. Still, the novelty claim is only moderately convincing because the method combines known ingredients rather than introducing a clearly new reconstruction principle, and the literature positioning is incomplete. The narrow baseline set and the citation inconsistency in Table 1 reduce confidence in the state-of-the-art claim.
- 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 is important and clinically relevant. A practical alternative to IOS from sparse photographs would have clear value for remote orthodontic workflows.
- The three-stage design is sensible. The decomposition into latent prior learning, conditional alignment, and mesh refinement is easy to follow and matches the geometry-centric task.
- The reported gains over the included baselines are sizable across several geometry metrics.
- The ablation study is useful and supports the contribution of cross-attention and counterfactual learning
- 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.
- Novelty is only moderate. The paper combines known components rather than clearly introducing a fundamentally new reconstruction paradigm, and the distinction from prior work such as TeethDreamer is not sharply articulated.
- The literature comparison is incomplete. A directly related 2025 sparse-view intraoral reconstruction paper is cited but not compared or meaningfully discussed: Dental3R.
- The experimental comparison is narrow for a strong SOTA claim, with only two baselines.
- The manuscript contains at least one factual/citation issue: the template-based baseline in Table 1 appears mis-cited as [11] instead of [1], where [1] is the actual five-photo parametric reconstruction method (source).
- The real clinical data description is too limited, and there is no visible ethics approval or consent statement for the 100 hospital cases.
- 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 mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- 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?
This paper addresses an important problem in digital dentistry and presents a technically coherent framework with a clear geometry-oriented design. The combination of latent geometric priors, conditional diffusion-based alignment, and statistical shape refinement is reasonable and appears effective in practice. The reported results are promising and suggest a meaningful improvement over the included baselines. At the same time, the paper does not yet fully establish a strong state-of-the-art contribution because the novelty over prior components is moderate, the baseline set is limited, and some methodological and data details remain insufficiently described. I therefore view the paper as marginally above the threshold: it has clear practical relevance and encouraging empirical performance, but it would benefit from stronger literature positioning, cleaner reporting, and clarification of the real-data and ethics details. My positive leaning depends in part on whether the rebuttal can address these concerns clearly.
- 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
Proposed ARCH-Net, a three-stage framework for reconstructing 3D dental models from five sparse intra-oral photographs (i) a VQ-VAE latent space trained on 3D tooth point clouds, (ii) conditional latent diffusion with cross-attention and a counterfactual symptom-aware contrastive objective for 2D–3D alignment, and (iii) statistical-shape-model-based mesh refinement. The problem is clinically relevant and the pipeline is coherent. The most distinctive part of the paper is the symptom-perturbation counterfactual supervision in Stage II. That said, the overall contribution reads more as a thoughtful integration of existing components than as a clearly transformative methodological advance.
- 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 paper addresses a meaningful clinical problem that is obtaining anatomically plausible 3D dental reconstructions from sparse intra-oral photographs could reduce dependence on expensive scanners and improve accessibility for remote orthodontic assessment. The clinical motivation is clear throughout the paper. 2.The decomposition into latent prior learning, conditional 2D–3D alignment, and shape-prior refinement is logically organized and easier to follow than many end-to-end black-box alternatives. 3.The paper also includes both ablation studies and an external real-image evaluation setting. In particular, the ablations on cross-attention and counterfactual learning indicate that both components matter within the proposed framework, and the inclusion of 100 hospital cases is compelling than a purely synthetic only evaluation.
- 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 provides nearly no detail on the 100 hospital cases, including inclusion criteria, patient variability, symptom distribution, imaging protocol, reference scan acquisition, and annotation/registration process. 2.The baseline comparison is too limited to support a state-of-the-art claim. Only two baselines are included, despite citing other recent and relevant dental reconstruction methods. A stronger comparison against more recent geometry-aware approaches is needed. 3.ARCH-Net uses SAM-derived tooth contours, while TeethDreamer operates directly on photographs. Since contour extraction may simplify the task, it is better separate the benefit of the input representation from the benefit of the proposed model. 3.The abstract and conclusion claim improved efficiency, robustness, and potential for real-time use, but no runtime, latency, memory, or robustness experiments are provided to support these statements. 4.Reported only aggregate metrics, without standard deviations, confidence intervals, case-level analysis, or significance testing. This made the reported improvements less convincing, especially where the margins are modest. 5.Important implementation details are missing, including symptom label construction, counterfactual sample generation, negative selection, loss weighting, SSM construction, PCA settings, and registration hyperparameters.
- 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.
- 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.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper presents an important and clinically relevant problem, and the overall framework is well motivated. I also appreciate that the authors didn’t rely on a purely unconstrained generative formulation and instead incorporated geometric priors, alignment supervision, and shape-model refinement. These are positive aspects. However, the experimental evidence is not yet strong enough for the level of claims being made. In addition, the paper lacks stronger and more current baseline comparisons, does not provide statistical significance or uncertainty analysis, and makes efficiency/robustness claims that are not experimentally demonstrated. Reproducibility is also limited by missing implementation and dataset details.
- 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 ARCH-Net, a three-stage hierarchical framework designed to reconstruct anatomically accurate 3D dental models from sparse 2D intra-oral photographs. To address the lack of geometric rigor in existing methods, the framework first learns a structured latent space by pretraining a VQ-VAE on 3D teeth point clouds. The core contribution is the second stage, where a latent diffusion model conditioned on 2D image contours bridges the 2D-3D gap. This diffusion process is supervised by a novel symptom-perturbation counterfactual contrastive loss to enforce rigorous geometric consistency for clinical features like overbite and malocclusion. Finally, the method utilizes Statistical Shape Models (SSM) as a prior to refine the generated coarse point clouds into high-resolution 3D meshes via rigid and non-rigid registration.
- 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 introduction of a counterfactual contrastive learning strategy during the latent diffusion process is a clever and highly relevant approach to enforcing clinical symptom consistency (e.g., overjet, overbite) in an inherently ill-posed 2D-to-3D reconstruction task.
- The framework naturally handles scenarios involving missing teeth caused by limited visibility in input photographs, an issue that causes significant artifacts in the compared baselines.
- The decomposition of the problem into an anatomically-constrained latent space construction, counterfactual-enhanced alignment, and prior-driven refinement systematically addresses the specific limitations of direct 2D-to-3D mapping strategies.
- The method demonstrates superior quantitative performance against state-of-the-art baselines across multiple rigorous geometric metrics, including Root Mean Squared Surface Distance (RMSD) and the Dice Similarity Coefficient (DSC).
- 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 model relies heavily on a synthetic dataset of 1,060 rendered contours for training, and its validation on real-world clinical data is limited to a relatively small sample size of 100 hospital cases.
- The framework heavily relies on the extraction of accurate teeth contours as inputs for the diffusion model. While the authors mention using the Segment Anything Model (SAM) to extract contours for baseline comparisons, the dependency on and potential compounding errors of this external segmentation step within their own pipeline are not thoroughly evaluated.
- The experimental comparison is restricted to only two baselines: one template-based method and one learning-based method (TeethDreamer). Missing comparisons with recent NeRF/3DGS-based reconstruction methods or strong multi-view or monocular 3D reconstruction baselines.
- No evaluation of clinical usability, orthodontist feedback or downstream tasks (e.g., treatment planning). Claims of “clinical applicability” are not substantiated.
- Possible typos: to decomposes the 3D teeth reconstruction task into three stages: —> (should be “decompose”)
- 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.
- 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 primary dataset uses rendered contours from 3D scans, not real intra-oral images. This creates a simulation gap and undermines claims of real-world applicability. The real dataset is very small (100 cases) and lacks detailed description (no diversity, acquisition protocol, or annotation details). The paper falls around the acceptance threshold.
- 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
N/A
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.
This paper, ARCH-Net, addresses the clinically important problem of reconstructing aligned 3D teeth models from partial intra-oral scans for orthodontic treatment planning. The core contribution is a hierarchical coarse-to-fine framework enhanced by a novel counterfactual contrastive learning module that improves alignment and reconstruction quality under challenging imaging conditions with partial occlusions and missing data.
The reviewers collectively rated the paper with an average score of 3.67, and after thorough evaluation, I recommend acceptance based on the following assessment:
Key Strengths Justifying Acceptance:
First, the counterfactual contrastive learning module represents a genuinely novel technical contribution. The integration of counterfactual reasoning into a hierarchical 3D dental reconstruction pipeline addresses an underexplored problem—learning robust feature representations invariant to partial occlusions and missing data. The reviewers consistently praised this component, with one reviewer specifically noting the “clever” nature of the approach.
Second, the ablation studies are thorough and effectively isolate the contribution of each proposed component. This level of experimental rigor is commendable and demonstrates a clear understanding of the methodological contribution. The reviewers unanimously acknowledged the quality of these ablation studies.
Third, the clinical relevance is unambiguous. The problem of accurate 3D teeth reconstruction from partial scans is a genuine unmet need in digital dentistry, with direct applications in orthodontic treatment planning and prosthodontics. Reviewer 2 highlighted the practical relevance, noting that the method addresses a real clinical challenge.
Addressing Remaining Concerns:
While the baseline comparison is limited to two methods (below the ideal standard for MICCAI), the comprehensive ablation studies and strong qualitative results partially compensate for this limitation. The paper’s strengths in novelty, ablation rigor, and clinical relevance outweigh the baseline comparison concern, which represents an incremental improvement opportunity rather than a fundamental flaw.
The ethical documentation concern regarding hospital data is valid but does not preclude acceptance. This is an administrative oversight that can be addressed in the camera-ready version with appropriate ethics statement inclusion.
Conclusion:
Given the genuine technical innovation, excellent ablation studies, clear clinical motivation, and positive reviewer assessments, this paper meets the threshold for MICCAI acceptance. The work represents a meaningful contribution to computational orthodontics and 3D dental reconstruction.