pith. sign in

arxiv: 2605.20827 · v1 · pith:G5IAGYG5new · submitted 2026-05-20 · 💻 cs.CV

HyDAR-Pano3D: A Hybrid Disentangled Anatomical Recovery Framework for Panoramic-to-3D Reconstruction

Yaoyao Yue , J\'er\^ome Schmid , Xiaoshuang Li , Eduardo Delamare , Jinman Kim This is my paper

Pith reviewed 2026-05-21 04:56 UTC · model grok-4.3

classification 💻 cs.CV
keywords panoramic radiographCBCT reconstructiondisentangled learning3D dental imagingdeformation fieldsemantic priorsanatomical recovery
0
0 comments X

The pith

HyDAR-Pano3D reconstructs accurate 3D dental volumes from 2D panoramic radiographs by first creating an arch-normalized canonical volume and then restoring individual features with a constrained deformation field.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that direct regression from panoramic radiographs to cone-beam CT volumes forces a model to learn both shared anatomy and patient-specific differences at once, which creates ambiguity and produces blurry results with unclear boundaries. It addresses this by splitting the task into two stages: the first builds a standardized canonical volume using radiographic features combined with semantic priors, and the second applies a structured deformation to bring the volume back to the patient's native shape. This disentangled approach matters for routine dental care because panoramic X-rays are common and cheap, yet 3D information would help with planning and assessment when full CT scans are unavailable. Experiments across three datasets confirm the volumes support accurate downstream tasks such as segmenting teeth and the inferior alveolar canal.

Core claim

The central claim is that reformulating panoramic-to-CBCT reconstruction as a disentangled anatomical recovery problem, with Stage 1 producing an arch-normalized canonical volume via dual-encoder integration of radiographic features and SAM-derived semantic priors and Stage 2 applying a prior-constrained structured deformation field via the Anatomical Restoration Network, reduces ambiguity in the inverse problem and yields higher-fidelity 3D volumes that preserve clinically relevant structures better than entangled direct-mapping baselines.

What carries the argument

The two-stage HyDAR-Pano3D framework consisting of a dual-encoder network that reconstructs the arch-normalized canonical volume and an Anatomical Restoration Network that predicts the prior-constrained structured deformation field.

If this is right

  • The synthesized volumes achieve 25.76 dB PSNR, 85.70 percent SSIM, and 83.83 percent overall anatomical Dice score while outperforming baselines at p less than 0.05.
  • Downstream segmentation of whole teeth reaches 82.4 percent Dice and inferior alveolar canal reaches 72.2 percent Dice using the reconstructed volumes.
  • The approach enables anatomy-aware assessment in settings where CBCT data cannot be acquired.
  • Individual morphological variations are restored without over-smoothing shared anatomical boundaries.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same separation of canonical form from patient-specific deformation could apply to other 2D-to-3D medical imaging tasks that involve roughly standardized anatomy.
  • Semantic priors from general vision models may reduce the need for large amounts of paired 2D-3D training data in specialized domains.
  • Testing the method on cases with pathologies or extreme anatomical variations would reveal how far the prior-constrained deformation can generalize.

Load-bearing premise

The assumption that SAM-derived semantic priors accurately capture dental anatomical structures in panoramic radiographs without introducing errors that propagate through the reconstruction stages.

What would settle it

If the final 3D volumes fail to improve tooth segmentation Dice scores above the reported 82.4 percent or show no statistically significant gain over direct-regression baselines on held-out panoramic radiographs, the disentanglement benefit would not hold.

Figures

Figures reproduced from arXiv: 2605.20827 by Eduardo Delamare, J\'er\^ome Schmid, Jinman Kim, Xiaoshuang Li, Yaoyao Yue.

Figure 1
Figure 1. Figure 1: HyDAR-Pano3D Overview. The framework factorizes 3D dental reconstruction into: (1) Canonical Volumetric Inference: Stage 1 maps 2D panoramic [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative comparison of panoramic 2D-to-3D dental reconstruction [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison of downstream tooth segmentation on the [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison of downstream tooth and inferior alveolar [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison of downstream tooth segmentation on the [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
read the original abstract

Panoramic radiograph (PR) is fundamentally used in routine dental care, but it inherently provides only a two-dimensional (2D) projection of complex three-dimensional (3D) craniofacial anatomy. Most existing learning-based methods attempt to computationally recover this 3D information by directly regressing native cone-beam computed tomography (CBCT) volumes from PR. However, this direct mapping requires the model to simultaneously learn common anatomical structures and patient-specific morphological variations. This entangled formulation makes the ill-posed 2D-to-3D inverse problem highly ambiguous, often producing over-smoothed reconstructions with blurred anatomical boundaries. To address this, we propose HyDAR-Pano3D, a two-stage framework that reformulates PR-to-CBCT reconstruction as a disentangled anatomical recovery problem. In Stage 1, a dual-encoder network integrates radiographic features with SAM-derived semantic priors to reconstruct an arch-normalized canonical volume. In Stage 2, an Anatomical Restoration Network predicts a prior-constrained structured deformation field to map this canonical volume back to the native space, restoring individual morphological variations. Experiments on three large-scale datasets show that HyDAR-Pano3D significantly outperforms baseline methods ($p < 0.05$), achieving a 25.76 dB PSNR, 85.70\% SSIM, and an 83.83\% overall anatomical Dice score. The synthesized volumes successfully support downstream segmentation of whole teeth (82.4\% Dice) and the inferior alveolar canal (72.2\% Dice), demonstrating that our disentangled approach preserves clinically relevant structures to enable robust anatomy-aware assessment when CBCT data is unavailable.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript presents HyDAR-Pano3D, a two-stage hybrid disentangled framework for panoramic radiograph (PR) to 3D CBCT reconstruction. Stage 1 integrates radiographic features with SAM-derived semantic priors via a dual-encoder network to produce an arch-normalized canonical volume; Stage 2 employs an Anatomical Restoration Network to predict a prior-constrained structured deformation field that restores patient-specific morphology. Experiments on three large-scale datasets report statistically significant gains over baselines (p < 0.05), with 25.76 dB PSNR, 85.70% SSIM, 83.83% overall anatomical Dice, and downstream segmentation Dice of 82.4% (whole teeth) and 72.2% (inferior alveolar canal).

Significance. If the results hold after addressing domain-shift concerns, the work could meaningfully advance dental imaging by enabling accurate 3D anatomical recovery from routine 2D PRs when CBCT is unavailable. The disentangled formulation directly targets the ambiguity of direct regression, and the reported downstream task utility indicates potential clinical value. The use of semantic priors and structured deformation is a clear strength for preserving boundaries.

major comments (2)
  1. [Stage 1 (abstract and §3)] Stage 1 description (abstract and method overview): The central performance claim (25.76 dB PSNR, 83.83% anatomical Dice) depends on SAM-derived semantic priors accurately delineating teeth, canals, and arch anatomy in PRs. No quantitative validation, error maps, or domain-adaptation analysis of these priors is provided; because SAM was trained on natural images, mis-segmentations on overlapping or low-contrast radiographic structures could corrupt the canonical volume and bias the Stage-2 deformation field.
  2. [Experiments] Experiments section (abstract): Statistically significant outperformance and downstream Dice scores are reported, yet no ablation studies isolate the contribution of the SAM priors or the two-stage disentanglement from other factors such as network capacity or training schedule. Without these controls it is difficult to attribute the gains specifically to the hybrid disentangled formulation rather than implementation details.
minor comments (2)
  1. [Abstract] The abstract states results on 'three large-scale datasets' but does not name them or summarize their characteristics (e.g., number of subjects, acquisition protocols), which would strengthen claims of generalizability.
  2. [Method] Notation for the 'prior-constrained structured deformation field' could be formalized with an explicit equation in the method section to improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and positive assessment of the potential clinical impact of the disentangled formulation. We address each major comment below and describe the revisions planned for the next manuscript version.

read point-by-point responses

  1. Referee: [Stage 1 (abstract and §3)] Stage 1 description (abstract and method overview): The central performance claim (25.76 dB PSNR, 83.83% anatomical Dice) depends on SAM-derived semantic priors accurately delineating teeth, canals, and arch anatomy in PRs. No quantitative validation, error maps, or domain-adaptation analysis of these priors is provided; because SAM was trained on natural images, mis-segmentations on overlapping or low-contrast radiographic structures could corrupt the canonical volume and bias the Stage-2 deformation field.

    Authors: We agree that quantitative validation of the SAM-derived priors on panoramic radiographs is necessary to support the performance claims. The manuscript describes their integration via the dual-encoder network but does not report domain-specific metrics such as segmentation accuracy, error maps, or adaptation analysis. In the revised version we will add a dedicated analysis subsection with Dice/IoU scores on a held-out PR subset, qualitative error visualizations, and discussion of the prompt-engineering steps used to adapt SAM to radiographic structures. This will directly address concerns about potential mis-segmentations and their effect on the canonical volume. revision: yes

  2. Referee: [Experiments] Experiments section (abstract): Statistically significant outperformance and downstream Dice scores are reported, yet no ablation studies isolate the contribution of the SAM priors or the two-stage disentanglement from other factors such as network capacity or training schedule. Without these controls it is difficult to attribute the gains specifically to the hybrid disentangled formulation rather than implementation details.

    Authors: We concur that ablation experiments are required to isolate the contribution of the SAM priors and the two-stage disentanglement. The current results show overall gains and downstream utility, yet lack explicit controls that remove the priors or compare against a capacity-matched single-stage regressor. We will perform and report these ablations in the revised manuscript, including (i) a variant without SAM priors and (ii) a direct single-stage baseline, with corresponding metric tables to quantify the incremental benefit of each component. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical validation stands independent of inputs

full rationale

The paper presents a two-stage neural framework (Stage 1: dual-encoder with SAM priors for canonical volume; Stage 2: prior-constrained deformation) whose performance is measured by direct experimental comparison to baselines on three datasets, yielding concrete metrics such as 25.76 dB PSNR and 83.83% Dice. No equations, fitted parameters renamed as predictions, or self-citation chains are invoked to derive the core claims; the disentanglement is an architectural choice justified by the reported outperformance rather than by construction from the inputs themselves. The derivation chain is therefore self-contained and externally falsifiable via the held-out test results.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The framework depends on a pre-trained SAM model for semantic priors and on learned network weights; the disentanglement itself is the main added element beyond standard supervised reconstruction.

free parameters (1)
  • network architecture hyperparameters and training weights
    The dual-encoder and deformation prediction networks contain numerous parameters fitted to the training data on the three datasets.
axioms (1)
  • domain assumption SAM provides accurate semantic priors for craniofacial structures in panoramic radiographs
    Invoked in Stage 1 to integrate with radiographic features for canonical volume reconstruction.

pith-pipeline@v0.9.0 · 5857 in / 1474 out tokens · 38405 ms · 2026-05-21T04:56:39.636026+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages

  1. [1]

    Patient risk related to common dental radiographic examinations: the impact of 2007 International Commission on Radiological Protection recommendations regarding dose calculation,

    J. B. Ludlow, L. E. Davies-Ludlow, and S. C. White, “Patient risk related to common dental radiographic examinations: the impact of 2007 International Commission on Radiological Protection recommendations regarding dose calculation,” J. Am. Dent. Assoc. , no. 9, pp. 1237–1243, 2008

    work page 2007

  2. [2]

    S. M. Mallya and E. W. N. Lam, White and Pharoah’s Oral Radiology: Principles and Interpretation , 8th ed. St. Louis, MO, USA: Elsevier, 2019

    work page 2019

  3. [3]

    Basic Knowledge and New Advances in Panoramic Radiography Imag- ing Techniques: A Narrative Review on What Dentists and Radiologists Should Know,

    R. Izzetti, M. Nisi, G. Aringhieri, L. Crocetti, F. Graziani, and C. Nardi, “Basic Knowledge and New Advances in Panoramic Radiography Imag- ing Techniques: A Narrative Review on What Dentists and Radiologists Should Know,” Appl. Sci., vol. 11, no. 17, p. 7858, 2021

    work page 2021

  4. [4]

    Den- tomaxillofacial imaging with panoramic views and cone beam CT,

    A. Suomalainen, E. Pakbaznejad Esmaeili, and S. Robinson, “Den- tomaxillofacial imaging with panoramic views and cone beam CT,” Insights Imaging, vol. 6, no. 1, pp. 1–16, 2015

    work page 2015

  5. [5]

    D. P. Antony, T. Thomas, and M. Nivedhitha, “Two-dimensional Peri- apical, Panoramic Radiography Versus Three-dimensional Cone-beam Computed Tomography in the Detection of Periapical Lesion After Endodontic Treatment: A Systematic Review,” Cureus, vol. 12, no. 4, p. e7736, 2020

    work page 2020

  6. [6]

    Comparison of the Efficacy of the Panoramic and Cone Beam Computed Tomography Imaging Methods in the Surgical Planning of the Maxillary All-On-4, M-4, and V-4,

    E. Tandogdu, A. Ayali, and M. G. Caymaz, “Comparison of the Efficacy of the Panoramic and Cone Beam Computed Tomography Imaging Methods in the Surgical Planning of the Maxillary All-On-4, M-4, and V-4,” BioMed Res. Int. , vol. 2022, p. 1553340, 2022

    work page 2022

  7. [7]

    Compari- son of panoramic radiography and CBCT to identify maxillary posterior roots invading the maxillary sinus,

    L. J. Lopes, T. O. Gamba, J. V . Bertinato, and D. Q. Freitas, “Compari- son of panoramic radiography and CBCT to identify maxillary posterior roots invading the maxillary sinus,” Dentomaxillofac. Radiol. , vol. 45, no. 6, p. 20160043, 2016

    work page 2016

  8. [8]

    Accuracy and reliability of cone-beam computed tomography for measuring alveolar bone height and detecting bony dehiscences and fenestrations,

    C. C. Leung, L. Palomo, R. Griffith, and M. G. Hans, “Accuracy and reliability of cone-beam computed tomography for measuring alveolar bone height and detecting bony dehiscences and fenestrations,” Am. J. Orthod. Dentofacial Orthop. , vol. 137, no. 4, pp. S109–S119, 2010

    work page 2010

  9. [9]

    Maxillofa- cial cone beam computed tomography: essence, elements and steps to interpretation,

    W. Scarfe, Z. Li, W. Aboelmaaty, S. Scott, and A. Farman, “Maxillofa- cial cone beam computed tomography: essence, elements and steps to interpretation,” Aust. Dent. J. , vol. 57, no. s1, pp. 46–60, 2012

    work page 2012

  10. [10]

    Radiation dose and accuracy analysis of newly developed cone-beam CT for dental and maxillofacial imaging,

    S. S. Thongvigitmanee, N. Pongnapang, S. Aootaphao, P. Yampri, T. Srivongsa, P. Sirisalee, J. Rajruangrabin, and P. Thajchayapong, “Radiation dose and accuracy analysis of newly developed cone-beam CT for dental and maxillofacial imaging,” in Proc. 35th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (EMBC) , Osaka, 2013, pp. 2356–2359

    work page 2013

  11. [11]

    Cone-beam computed tomography: Time to move from ALARA to ALADA,

    P. P. Jaju and S. P. Jaju, “Cone-beam computed tomography: Time to move from ALARA to ALADA,” Imaging Sci. Dent. , vol. 45, no. 4, p. 263, 2015

    work page 2015

  12. [12]

    Oral-3D: Recon- structing the 3D structure of oral cavity from panoramic x-ray,

    W. Song, Y . Liang, J. Yang, K. Wang, and L. He, “Oral-3D: Recon- structing the 3D structure of oral cavity from panoramic x-ray,” in Proc. AAAI Conf. Artif. Intell. , vol. 35, 2021, pp. 566–573

    work page 2021

  13. [13]

    PX2Tooth: Recon- structing the 3D Point Cloud Teeth from a Single Panoramic X-Ray,

    W. Ma, H. Wu, Z. Xiao, Y . Feng, J. Wu, and Z. Liu, “PX2Tooth: Recon- structing the 3D Point Cloud Teeth from a Single Panoramic X-Ray,” in Proc. Med. Image Comput. Comput. Assist. Intervent. (MICCAI) , vol. 15003. Springer, 2024, pp. 411–421

    work page 2024

  14. [14]

    ViT- NeBLa: A hybrid vision transformer and neural beer–lambert framework for single-view 3D reconstruction of oral anatomy from panoramic radiographs,

    B. Keshari Parida, A. P. Sunilkumar, A. Sen, and W. You, “ViT- NeBLa: A hybrid vision transformer and neural beer–lambert framework for single-view 3D reconstruction of oral anatomy from panoramic radiographs,” IEEE Access, vol. 13, pp. 170 761–170 781, 2025

    work page 2025

  15. [15]

    3D reconstruction from 2D multi-view dental images based on EfficientNetB0 model,

    W. Mohamed, N. Nader, Y . M. Alsakar, N. Elazab, M. Ezzat, and M. Elmogy, “3D reconstruction from 2D multi-view dental images based on EfficientNetB0 model,” Sci. Rep., vol. 15, no. 1, p. 28775, 2025

    work page 2025

  16. [16]

    What is Cone-Beam CT and How Does it Work?

    W. C. Scarfe and A. G. Farman, “What is Cone-Beam CT and How Does it Work?” Dent. Clin. North Am. , vol. 52, no. 4, pp. 707–730, 2008

    work page 2008

  17. [17]

    X2CT-GAN: reconstructing CT from biplanar X-rays with generative adversarial networks,

    X. Ying, H. Guo, K. Ma, J. Wu, Z. Weng, and Y . Zheng, “X2CT-GAN: reconstructing CT from biplanar X-rays with generative adversarial networks,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2019, pp. 10 619–10 628

    work page 2019

  18. [18]

    MedNeRF: Medical neural radiance fields for reconstructing 3D-aware CT-projections from a single X-ray,

    A. Corona-Figueroa, J. Frawley, S. B. Taylor, S. Bethapudi, H. P. H. Shum, and C. G. Willcocks, “MedNeRF: Medical neural radiance fields for reconstructing 3D-aware CT-projections from a single X-ray,” in Proc. 44th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. (EMBC) , 2022, pp. 3843–3848

    work page 2022

  19. [19]

    Conditional-Flow NeRF: Accurate 3D Modelling with Reliable Uncertainty Quantifica- tion,

    J. Shen, A. Agudo, F. Moreno-Noguer, and A. Ruiz, “Conditional-Flow NeRF: Accurate 3D Modelling with Reliable Uncertainty Quantifica- tion,” in Proc. Comput. Vis. – ECCV , vol. 13663. Springer, 2022, pp. 540–557

    work page 2022

  20. [20]

    DTR-Net: Dual-Space 3D Tooth Model Reconstruction From Panoramic X-Ray Images,

    L. Mei, Y . Fang, Y . Zhao, X. S. Zhou, M. Zhu, Z. Cui, and D. Shen, “DTR-Net: Dual-Space 3D Tooth Model Reconstruction From Panoramic X-Ray Images,” IEEE Trans. Med. Imaging , vol. 43, no. 1, pp. 517–528, 2024

    work page 2024

  21. [21]

    X2Teeth: 3D Teeth Reconstruction from a Single Panoramic Radiograph,

    Y . Liang, W. Song, J. Yang, L. Qiu, K. Wang, and L. He, “X2Teeth: 3D Teeth Reconstruction from a Single Panoramic Radiograph,” in Proc. Med. Image Comput. Comput. Assist. Intervent. (MICCAI) , vol. 12262. Springer, 2020, pp. 400–409

    work page 2020

  22. [22]

    3D Teeth Reconstruction from Panoramic Radiographs Using Neural Implicit Functions,

    S. Park, S. Kim, I.-S. Song, and S. J. Baek, “3D Teeth Reconstruction from Panoramic Radiographs Using Neural Implicit Functions,” in Lect. Notes Comput. Sci. , vol. 14229. Springer, 2023, pp. 376–386

    work page 2023

  23. [23]

    V oxelmorph: a learning framework for deformable medical image registration,

    G. Balakrishnan, A. Zhao, M. R. Sabuncu, J. Guttag, and A. V . Dalca, “V oxelmorph: a learning framework for deformable medical image registration,” IEEE Trans. Med. Imaging, vol. 38, no. 8, pp. 1788–1800, 2019

    work page 2019

  24. [24]

    Deep learning techniques for inverse problems in imaging,

    G. Ongie, A. Jalal, C. A. Metzler, R. G. Baraniuk, A. G. Dimakis, and R. Willett, “Deep learning techniques for inverse problems in imaging,” IEEE J. Sel. Areas Inf. Theory , vol. 1, no. 1, pp. 39–56, 2020. 10

    work page 2020

  25. [25]

    On instabilities of deep learning in image reconstruction and the potential costs of AI,

    V . Antun, F. Renna, C. Poon, B. Adcock, and A. C. Hansen, “On instabilities of deep learning in image reconstruction and the potential costs of AI,” Proc. Nat. Acad. Sci. , vol. 117, no. 48, pp. 30 088–30 095, 2020

    work page 2020

  26. [26]

    Considering anatomical prior information for low-dose CT image enhancement using attribute- augmented Wasserstein generative adversarial networks,

    Z. Huang, X. Liu, R. Wang, J. Chen, P. Lu, Q. Zhang, C. Jiang, Y . Yang, X. Liu, H. Zheng, D. Liang, and Z. Hu, “Considering anatomical prior information for low-dose CT image enhancement using attribute- augmented Wasserstein generative adversarial networks,” Neurocomput- ing, vol. 428, pp. 104–115, 2021

    work page 2021

  27. [27]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, and W.-Y . Lo, “Segment anything,” in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV) , 2023, pp. 4015– 4026

    work page 2023

  28. [28]

    Segment anything in medical images,

    J. Ma, Y . He, F. Li, L. Han, C. You, and B. Wang, “Segment anything in medical images,” Nat. Commun., vol. 15, no. 1, p. 654, 2024

    work page 2024

  29. [29]

    SemiT-SAM: Building A Visual Foundation Model for Tooth Instance Segmentation on Panoramic Radiographs,

    J. Hao, M. Liu, L. He, L. Yao, J. K. H. Tsoi, and K. F. Hung, “SemiT-SAM: Building A Visual Foundation Model for Tooth Instance Segmentation on Panoramic Radiographs,” in Proc. Supervised Semi- Supervised Multi-Struct. Segment. Landmark Detect. Dental Data , vol. 15571. Springer, 2025, pp. 110–121

    work page 2025

  30. [30]

    Segmentation-guided MRI reconstruction for meaningfully diverse reconstructions,

    J. N. Morshuis, M. Hein, and C. F. Baumgartner, “Segmentation-guided MRI reconstruction for meaningfully diverse reconstructions,” in Proc. MICCAI Workshop Deep Gen. Models , 2024, pp. 180–190

    work page 2024

  31. [31]

    On the Utility of Foundation Models for Fast MRI: Vision-Language-Guided Image Reconstruction,

    R. Feng, X. He, R. Mercer, Z. Stewart, and F. Liu, “On the Utility of Foundation Models for Fast MRI: Vision-Language-Guided Image Reconstruction,” arXiv preprint arXiv:2511.19641 , 2025

    work page arXiv 2025

  32. [32]

    3DPX: Progressive 2D-to-3D Oral Image Reconstruction with Hybrid MLP-CNN Networks,

    X. Li, M. Meng, Z. Huang, L. Bi, E. Delamare, D. Feng, B. Sheng, and J. Kim, “3DPX: Progressive 2D-to-3D Oral Image Reconstruction with Hybrid MLP-CNN Networks,” in Proc. Med. Image Comput. Comput. Assist. Intervent. (MICCAI) , vol. 15007. Springer, 2024, pp. 25–34

    work page 2024

  33. [33]

    Free-form deformation of solid geometric models,

    T. W. Sederberg and S. R. Parry, “Free-form deformation of solid geometric models,” SIGGRAPH Comput. Graph. , vol. 20, no. 10, pp. 151–160, 1986

    work page 1986

  34. [34]

    CPR - curved planar reformation,

    A. Kanitsar, D. Fleischmann, R. Wegenkittl, P. Felkel, and E. Groller, “CPR - curved planar reformation,” in Proc. IEEE Visualization (VIS) . IEEE, 2002, pp. 37–44

    work page 2002

  35. [35]

    Nonrigid registration using free-form deformations: application to breast MR images,

    D. Rueckert, L. Sonoda, C. Hayes, D. Hill, M. Leach, and D. Hawkes, “Nonrigid registration using free-form deformations: application to breast MR images,” IEEE Trans. Med. Imaging, vol. 18, no. 8, pp. 712– 721, 1999

    work page 1999

  36. [36]

    Mathematical analyses of dental arch curvature in normal occlusion,

    S. AlHarbi, E. A. Alkofide, and A. AlMadi, “Mathematical analyses of dental arch curvature in normal occlusion,” Angle Orthod. , vol. 78, no. 2, pp. 281–287, 2008

    work page 2008

  37. [37]

    The curve of dental arch in normal occlusion,

    A.-H. Muhamad, W. Nezar, and A. Azzaldeen, “The curve of dental arch in normal occlusion,” Open Sci. J. Clin. Med. , vol. 3, no. 2, pp. 47–54, 2015

    work page 2015

  38. [38]

    A fully automatic AI system for tooth and alveolar bone segmentation from cone-beam CT images,

    Z. Cui, Y . Fang, L. Mei, B. Zhang, B. Yu, J. Liu, C. Jiang, Y . Sun, L. Ma, and J. Huang, “A fully automatic AI system for tooth and alveolar bone segmentation from cone-beam CT images,” Nat. Commun., vol. 13, no. 1, p. 2096, 2022

    work page 2096

  39. [39]

    MMDental-A multimodal dataset of tooth CBCT images with expert medical records,

    C. Wang, Y . Zhang, C. Wu, J. Liu, X. Huang, L. Wu, Y . Wang, X. Feng, Y . Lu, and Y . Wang, “MMDental-A multimodal dataset of tooth CBCT images with expert medical records,” Sci. Data, vol. 12, no. 1, p. 1172, 2025

    work page 2025

  40. [40]

    Enhancing patch-based learning for the segmentation of the mandibular canal,

    L. Lumetti, V . Pipoli, F. Bolelli, E. Ficarra, and C. Grana, “Enhancing patch-based learning for the segmentation of the mandibular canal,” IEEE Access, vol. 12, pp. 79 014–79 024, 2024

    work page 2024

  41. [41]

    Segmenting the inferior alveolar canal in CBCTs volumes: the toothfairy challenge,

    F. Bolelli, L. Lumetti, S. Vinayahalingam, M. Di Bartolomeo, A. Pel- lacani, K. Marchesini, N. Van Nistelrooij, P. Van Lierop, T. Xi, Y . Liu et al. , “Segmenting the inferior alveolar canal in CBCTs volumes: the toothfairy challenge,” IEEE Trans. Med. Imaging , vol. 44, no. 4, pp. 1890–1906, 2024

    work page 1906

  42. [42]

    Im- proving segmentation of the inferior alveolar nerve through deep label propagation,

    M. Cipriano, S. Allegretti, F. Bolelli, F. Pollastri, and C. Grana, “Im- proving segmentation of the inferior alveolar nerve through deep label propagation,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2022, pp. 21 137–21 146

    work page 2022

  43. [43]

    UNETR: Transformers for 3D medical image segmentation,

    A. Hatamizadeh, Y . Tang, V . Nath, D. Yang, A. Myronenko, B. Land- man, H. R. Roth, and D. Xu, “UNETR: Transformers for 3D medical image segmentation,” in Proc. IEEE/CVF Winter Conf. Appl. Comput. Vis. (WACV), 2022, pp. 574–584

    work page 2022

  44. [44]

    Panoramic dental tomosynthesis imaging by use of CBCT projection data,

    T. Kwon, D.-i. Choi, J. Hwang, T. Lee, I. Lee, and S. Cho, “Panoramic dental tomosynthesis imaging by use of CBCT projection data,” Sci. Rep., vol. 13, no. 1, p. 8817, 2023

    work page 2023