pith. sign in

arxiv: 2605.15241 · v1 · pith:7LBFXLEDnew · submitted 2026-05-14 · 📡 eess.IV · cs.CV· cs.LG

From Full and Partial Intraoral Scans to Crown Proposal: A Classification-Guided Restoration Assistance Pipeline

Rabin Kunwar , Dikshya Parajuli , Rujal Acharya , Romik Gosai , Prince Panta , Kundan Siwakoti , Shuvangi Adhikari , Saugat Kafley
show 7 more authors
Louis Digiorgio Amit Regmi Akio Tanaka Masahiko Inada Yuriko Komagamine Kennta Kashiwazaki Manabu Kanazawa
This is my paper

Pith reviewed 2026-05-19 16:10 UTC · model grok-4.3

classification 📡 eess.IV cs.CVcs.LG
keywords intraoral scanscrown restorationdental segmentationpartial scans3D point cloudsDGCNNcrown proposal
0
0 comments X

The pith

A classification-guided pipeline produces patient-specific crown proposals from partial intraoral scans in 2.5-3.5 minutes.

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

The paper develops an end-to-end pipeline that accepts a raw intraoral scan and target tooth number to output an initial crown proposal for refinement. It routes partial-scan segmentation through a DGCNN classifier that identifies one of five anatomical types, then applies coarse-to-fine registration and graph-cut boundary refinement to achieve accurate tooth and gingiva labels. Context-aware retrieval then selects crown candidates by cosine similarity on DGCNN embeddings of neighboring and opposing teeth, followed by spline alignment and Blender fitting to produce the shell. This retrieval-based route is positioned as a practical alternative to end-to-end generative crown methods that tend to smooth occlusal details.

Core claim

The pipeline takes a raw intraoral scan and FDI tooth number as input and outputs an initial crown proposal. By routing segmentation through a classify-then-align strategy for partial scans and using DGCNN-based retrieval over neighboring teeth embeddings, it achieves macro-average DSC 0.9249 across 17 classes on 1,958 partial scans and generates the crown shell in 2.5-3.5 minutes.

What carries the argument

The classify-then-align segmentation strategy that first categorizes the scan into one of five anatomical types with DGCNN, applies coarse-to-fine RANSAC+ICP registration, and refines boundaries with graph-cut, followed by the context-aware retrieval using cosine similarity on DGCNN embeddings of neighboring and opposing teeth.

If this is right

  • Partial scans achieve macro-average DSC 0.9249 with sub-millimeter centroid errors of 0.2666-0.2774 mm for the prepared tooth and neighbors.
  • The pipeline supplies a fast, practical alternative to generative approaches that lose occlusal detail.
  • High segmentation precision and recall support reliable downstream crown fitting in CAD/CAM workflows.
  • The method works on the partial scans preferred for single-unit cases to avoid stitching errors.

Where Pith is reading between the lines

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

  • The embedding-retrieval approach could extend to other single-tooth restorations such as inlays or onlays.
  • Adding a lightweight generative step after retrieval might further refine occlusal surfaces while retaining speed.
  • Expanding the candidate library or testing on underrepresented anatomies would clarify limits of the similarity-based selection.

Load-bearing premise

Cosine similarity over DGCNN embeddings of neighboring and opposing teeth will surface geometrically suitable crown candidates that match the specific patient anatomy.

What would settle it

A test set of partial scans where retrieved crowns repeatedly fail to match occlusal contacts or proximal surfaces accurately, forcing extensive manual edits, would show the retrieval step does not deliver clinically usable proposals.

Figures

Figures reproduced from arXiv: 2605.15241 by Akio Tanaka, Amit Regmi, Dikshya Parajuli, Kennta Kashiwazaki, Kundan Siwakoti, Louis Digiorgio, Manabu Kanazawa, Masahiko Inada, Prince Panta, Rabin Kunwar, Romik Gosai, Rujal Acharya, Saugat Kafley, Shuvangi Adhikari, Yuriko Komagamine.

Figure 1
Figure 1. Figure 1: Overview of the proposed crown generation pipeline. The dentist first specifies the target FDI tooth number as user input. (a) Raw intraoral scans are classified and aligned using DGCNN + RANSAC/ICP. (b) A segmentation network isolates individual tooth surfaces. (c) A neighbour-context-aware retrieval module uses the FDI number to query the crown library and selects the most morphologically similar shell. … view at source ↗
Figure 2
Figure 2. Figure 2: Ground Truth Annotation Schema. Visualization of the semantic labeling strategy based on the FDI notation system. (Top) The Maxillary arch, containing indices 11–18 (upper right) and 21–28 (upper left). (Bottom) The Mandibular arch, containing indices 31–38 (lower left) and 41–48 (lower right). Each FDI label is associated with a unique color used during the semantic segmentation training process. 2) Datas… view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of full-jaw tooth classes highlighting wisdom teeth. Red [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Preprocessing stage (i): Jaws classification of intraoral scans. registration. As shown in Table II, the model demonstrated exceptional reliability for lateral partial scans, with F1-scores of 0.944 (Left) and 0.957 (Right). High precision in these categories is vital because partial scans lack the global arch context required for standard alignment. TABLE II TEST SET CLASSIFICATION METRICS (8-FEATURE BEST… view at source ↗
Figure 6
Figure 6. Figure 6: Population Analysis. (Top) Visualization of all individual tooth centroids after global alignment. (Bottom) The computed “Global Average Centroid Curve” representing the population mean. dataset, we applied the orientation normalization pipeline described in [27]. This process aligns the occlusal plane to the global XY-plane and corrects anterior-posterior ambiguity using parabolic fitting, ensuring all 98… view at source ↗
Figure 7
Figure 7. Figure 7: Canonical Reference Templates. (a) The Canonical Full Upper and Lower jaws selected based on the best fit to the average curve. (b, c) The dictionary of Upper and Lower partial templates derived by cropping the Master templates, shown aligned to the global curve. Furthermore, to prevent misalignment errors along the den￾tal arch, we tightened the RANSAC geometric constraints by setting the Edge Length Simi… view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative registration validation on full and partial intraoral scans. The top row shows arbitrarily oriented input scans overlaid with the corresponding reference template, while the bottom row shows the final canonical alignment produced by the proposed Coarse-to-Fine registration pipeline. Full-arch examples (upper and lower) and partial-arch examples (left, right, and anterior) are presented together… view at source ↗
Figure 9
Figure 9. Figure 9: Library Crown Pre-annotation. Color-coded regions define the mesial (green), buccal (red), and occlusal (blue) faces. preparation site, Cprep. At this point on the curve, we calculate two initial direction vectors within the XY plane: • Reference Mesial Vector (vm−ref ): The normalized tangent to the curve, pointing “forward” (distal-to￾mesial) along the arch. • Reference Buccal Vector (vb−ref ): The norma… view at source ↗
Figure 10
Figure 10. Figure 10: Robust target extraction at the preparation site. (a) Mesial alignment: (left) full view showing orange and purple scatter of mesial vertices before and after alignment; (right) zoomed detail of the recovered target vector vm-robust at the prepared tooth. (b) Buccal alignment: similarly showing the recovered buccal target vb-robust. iprep}. 2. A Robust Mesial Target (vm-robust) is found by averag￾ing the … view at source ↗
Figure 11
Figure 11. Figure 11: Visual overview of the Patient-Specific Crown Alignment pipeline. The figure illustrates the process in three stages: (a) Initial Misalignment: The retrieved library crown (yellow) is initially positioned far from the patient scan (Grey). (b) Dental Arc: a cubic spline is fitted to the dental arch tooth centroids. At the preparation site, robust Mesial and Buccal target vectors are extracted. (c) Final Re… view at source ↗
Figure 12
Figure 12. Figure 12: Identification of the top five cusps used for local “tap-down” [PITH_FULL_IMAGE:figures/full_fig_p010_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Qualitative analysis of the five lowest scoring scans. (Top row) Ground truth annotations. (Bottom row) Model predictions. Per-case metrics are reported below each scan: Jaw DSC denotes the overall macro-average arch DSC; Ctx Dice and Ctx CE denote the macro-average DSC and Centroid Error (mm) computed over the context area, defined as the prepared tooth and its immediate mesial (Adjacent 1) and distal (A… view at source ↗
Figure 14
Figure 14. Figure 14: Qualitative evaluation of the crown-generation workflow on challenging posterior partial scan cases identified previously in [PITH_FULL_IMAGE:figures/full_fig_p014_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Misclassification impact on segmentation for a left-partial input. The same input jaw is aligned with each candidate template (full, central-partial, partial-left, and partial-right), then passed through the same segmentation model. Compatible template alignments (b–d) preserve high DSC, while contralateral partial-right alignment (e) yields catastrophic failure. [4] A. Ender and A. Mehl, “Effect of intra… view at source ↗
read the original abstract

Single-unit crown restoration is among the most common procedures in clinical dentistry, with CAD/CAM workflows now designing crowns directly from intraoral scans. Partial scans are often preferred over full-arch scans for single-unit cases due to fewer stitching errors, yet most segmentation networks trained on full arches fail on partial scans, while end-to-end generative crown methods often produce over-smoothed surfaces that lose occlusal detail. We propose an end-to-end pipeline that takes a raw intraoral scan and target FDI tooth number as input and outputs an initial, patient-specific crown proposal for clinician refinement. The pipeline has three phases: (I) data preparation and pose standardization; (II) segmentation routed by scan type; and (III) crown proposal generation via context-aware retrieval and Blender-based fitting. We address partial-scan segmentation through a classify-then-align strategy: a DGCNN classifier categorizes the scan into one of five anatomical types, then coarse-to-fine RANSAC+ICP registration standardizes the jaw coordinate frame, followed by graph-cut optimization to refine tooth-gingival boundaries. Trained on 1,958 partial scans, the pipeline achieves macro-average DSC 0.9249, Recall 0.8919, and Precision 0.9615 across 17 semantic classes; a fine-tuned full-arch model reaches DSC 0.9347. The prepared tooth and its mesial and distal neighbors achieve DSC 0.9468-0.9569 with sub-millimeter Centroid Errors (0.2666-0.2774 mm). These centroids anchor a retrieval module using DGCNN embeddings and cosine similarity over neighboring and opposing teeth, followed by spline-guided alignment and Blender Python API refinement. The pipeline produces a preliminary crown shell in 2.5-3.5 minutes, offering a practical alternative to end-to-end generative approaches.

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 proposes an end-to-end pipeline that accepts a raw intraoral scan and target FDI tooth number and outputs a preliminary patient-specific crown shell. Phase I standardizes pose; Phase II classifies the scan into one of five anatomical types with DGCNN, applies coarse-to-fine RANSAC+ICP registration, and refines tooth-gingival boundaries via graph-cut; Phase III retrieves library crowns via cosine similarity on DGCNN embeddings of neighboring and opposing teeth, then performs spline-guided alignment and Blender Python API fitting. On 1,958 partial scans the segmentation stage reports macro-average DSC 0.9249, recall 0.8919, precision 0.9615 across 17 classes, with prepared-tooth and neighbor DSC 0.9468-0.9569 and centroid errors 0.2666-0.2774 mm; total runtime is stated as 2.5-3.5 minutes.

Significance. The segmentation results on partial scans are quantitatively grounded and address a documented clinical pain point where full-arch models degrade. The classify-then-align strategy and reported sub-millimeter centroid accuracy constitute a concrete, reproducible contribution. If the retrieval step can be shown to select anatomically suitable crowns, the pipeline would supply a practical, controllable alternative to end-to-end generative crown methods; the absence of any surface-distance, occlusion, or expert-comparison metrics for the final proposals, however, prevents a full assessment of clinical utility.

major comments (2)
  1. [Abstract / crown proposal generation] Abstract and crown-proposal section: no quantitative evaluation (surface-to-surface distance, Hausdorff distance, occlusion contact area, or expert rating versus ground-truth clinical crowns) is supplied for the output of the DGCNN-embedding retrieval, spline alignment, and Blender fitting steps. Because the headline claim of a “clinically usable preliminary crown shell” rests on the premise that cosine similarity over neighboring/opposing embeddings surfaces appropriate library candidates, this omission is load-bearing for the central contribution.
  2. [Abstract] Abstract: the claim that the pipeline offers a “practical alternative to end-to-end generative approaches” is not supported by any baseline comparison or ablation that isolates the contribution of the context-aware retrieval module versus a naïve nearest-neighbor or random selection from the library.
minor comments (2)
  1. The manuscript should report the size and diversity of the crown library used for retrieval and any inclusion/exclusion criteria applied to the 1,958-scan training set.
  2. Train/test split details, cross-validation strategy, and whether the 1,958 partial scans are patient-disjoint should be stated explicitly.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive feedback and for recognizing the quantitative grounding of the segmentation results on partial scans. We address the two major comments below regarding the crown-proposal stage. Where the comments correctly identify gaps in the current evaluation, we commit to revisions that strengthen the manuscript without overstating what has been demonstrated.

read point-by-point responses

  1. Referee: [Abstract / crown proposal generation] Abstract and crown-proposal section: no quantitative evaluation (surface-to-surface distance, Hausdorff distance, occlusion contact area, or expert rating versus ground-truth clinical crowns) is supplied for the output of the DGCNN-embedding retrieval, spline alignment, and Blender fitting steps. Because the headline claim of a “clinically usable preliminary crown shell” rests on the premise that cosine similarity over neighboring/opposing embeddings surfaces appropriate library candidates, this omission is load-bearing for the central contribution.

    Authors: We agree that the absence of surface-distance and occlusion metrics for the retrieved and fitted crowns limits assessment of the final output. The manuscript currently demonstrates the retrieval and fitting pipeline through the embedding similarity mechanism and qualitative examples, while providing detailed quantitative results only for the upstream segmentation. In the revised version we will add surface-to-surface distance and Hausdorff distance metrics computed against available ground-truth clinical crowns on a subset of cases. Occlusion contact area and formal expert ratings would require a separate clinical validation protocol; we will explicitly note this as a limitation and future direction rather than claim full clinical usability at present. revision: partial

  2. Referee: [Abstract] Abstract: the claim that the pipeline offers a “practical alternative to end-to-end generative approaches” is not supported by any baseline comparison or ablation that isolates the contribution of the context-aware retrieval module versus a naïve nearest-neighbor or random selection from the library.

    Authors: We accept that an explicit ablation would better isolate the benefit of context-aware retrieval over simpler baselines. The current text motivates the approach via the use of neighboring and opposing tooth embeddings but does not report a direct comparison. In the revision we will insert a quantitative ablation that compares the proposed cosine-similarity retrieval against both random library selection and nearest-neighbor selection in the same embedding space, using embedding similarity scores and qualitative fit assessment as proxies. This addition will support the claim more rigorously while remaining within the scope of the existing library and embedding model. revision: yes

standing simulated objections not resolved
  • Formal expert clinical ratings of the crown proposals, which would require a new IRB-approved reader study outside the current dataset and experimental design.

Circularity Check

0 steps flagged

No circularity: empirical metrics on held-out scans and standard retrieval components do not reduce to inputs by construction

full rationale

The paper's central results are measured performance numbers (macro DSC 0.9249, centroid errors 0.2666-0.2774 mm) obtained by training and evaluating DGCNN and registration modules on 1,958 held-out partial scans. The crown proposal step invokes DGCNN embeddings + cosine similarity for retrieval followed by spline alignment and Blender fitting; these are methodological choices whose clinical suitability is an external assumption, not a quantity defined by the paper's own equations or prior self-citations. No self-definitional loops, fitted-input-as-prediction, or load-bearing self-citations appear in the derivation chain. The pipeline is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on trained DGCNN models and an implicit database of tooth geometries; no new physical entities are postulated. The main unstated premises are that the five anatomical types are sufficient to route all clinical partial scans and that embedding similarity correlates with crown-fit suitability.

free parameters (1)
  • Number of anatomical scan types
    Fixed at five to enable classify-then-align routing; chosen by the authors rather than derived.
axioms (2)
  • domain assumption DGCNN produces embeddings that capture clinically relevant tooth geometry for retrieval
    Invoked without further justification in the crown proposal generation phase.
  • domain assumption RANSAC+ICP registration followed by graph-cut yields accurate tooth-gingival boundaries on partial scans
    Core step in phase II pose standardization and boundary refinement.

pith-pipeline@v0.9.0 · 5952 in / 1712 out tokens · 74338 ms · 2026-05-19T16:10:27.460117+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

32 extracted references · 32 canonical work pages · 2 internal anchors

  1. [1]

    The transformative role of artificial intelligence in dentistry: A comprehensive overview. Part 1: Fundamentals of AI, and its contemporary applications in dentistry,

    L. Samaranayake, N. Tuygunov, F. Schwendicke, T. Osathanon, Z. Khurshid, S. A. Boymuradov, and A. Cahyanto, “The transformative role of artificial intelligence in dentistry: A comprehensive overview. Part 1: Fundamentals of AI, and its contemporary applications in dentistry,”International Dental Journal, vol. 75, no. 2, pp. 383– 396, 2025. [Online]. Avail...

    work page 2025

  2. [2]

    Dilated dynamic graph CNN for efficient point cloud learning,

    Y . Xu, X. Li, C. Wang, and S. Wen, “Dilated dynamic graph CNN for efficient point cloud learning,”IEEE Transactions on Image Processing, vol. 31, pp. 4468–4481, 2022. [Online]. Available: https://link.springer.com/article/10.1007/s10278-024-01061-6

    work page doi:10.1007/s10278-024-01061-6 2022

  3. [3]

    A critical analysis of the limitation of deep learning based 3D dental mesh segmentation methods in segmenting partial scans,

    A. Jana, A. Maiti, and D. N. Metaxas, “A critical analysis of the limitation of deep learning based 3D dental mesh segmentation methods in segmenting partial scans,” in2023 45th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), 2023, accepted to IEEE EMBC 2023. [Online]. Available: https://arxiv.org/abs/2305.002...

    work page arXiv 2023

  4. [4]

    Effect of intraoral scanner and fixed partial denture situation on the scan accuracy of multiple implants: an in vitro study,

    A. Ender and A. Mehl, “Effect of intraoral scanner and fixed partial denture situation on the scan accuracy of multiple implants: an in vitro study,”Clinical Oral Implants Research, vol. 27, no. 10, pp. 1309–1316, 2016

    work page 2016

  5. [5]

    Effect of scanned area and operator on the accuracy of dentate arch scans with a single implant,

    J.-M. Park, “Effect of scanned area and operator on the accuracy of dentate arch scans with a single implant,”The Journal of Prosthetic Dentistry, vol. 116, no. 4, pp. 534–540, 2016

    work page 2016

  6. [6]

    Accuracy of four intraoral scanners in oral implantology: a comparative in vitro study,

    M. Imburgia, S. Logozzo, U. Hauschild, G. Veronesi, C. Mangano, and F. G. Mangano, “Accuracy of four intraoral scanners in oral implantology: a comparative in vitro study,”BMC Oral Health, vol. 17, no. 1, p. 92, 2017

    work page 2017

  7. [7]

    Trueness and precision of 5 intraoral scanners in the impressions of single and multiple implants: a comparative in vitro study,

    F. G. Mangano, G. Veronesi, U. Hauschild, E. Mijiritsky, and C. Mangano, “Trueness and precision of 5 intraoral scanners in the impressions of single and multiple implants: a comparative in vitro study,”BMC Oral Health, vol. 19, no. 1, p. 101, 2019

    work page 2019

  8. [8]

    MeshSNet: Deep multi-scale mesh feature learning for end-to-end tooth segmentation,

    C. Lian, L. Wang, T.-H. Wu, M. Liu, F. Durran, D. Chambers, D. Wang, and D. Shen, “MeshSNet: Deep multi-scale mesh feature learning for end-to-end tooth segmentation,”IEEE Transactions on Medical Imaging, vol. 39, no. 11, pp. 3635–3645, 2020

    work page 2020

  9. [9]

    DArch: Dental arch prior-assisted 3D tooth instance segmentation,

    L. Qiu, Y . Zhang, E. Adeli, Q. Yu, K. Yan, G. Wang, X. Liu, L. Lu, D. Xu, and X. Xiang, “DArch: Dental arch prior-assisted 3D tooth instance segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 20 718– 20 727

    work page 2022

  10. [10]

    Dynamic Graph CNN for Learning on Point Clouds

    Y . Wang and J. M. Solomon, “Dynamic graph CNN for learning on point clouds,” inProceedings of the ACM SIGGRAPH Asia Conference. ACM, 2019. [Online]. Available: https://arxiv.org/abs/1801.07829

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  11. [11]

    PointNet++: Deep hierarchical feature learning on point sets in a metric space,

    C. R. Qi, L. Yi, H. Su, and L. J. Guibas, “PointNet++: Deep hierarchical feature learning on point sets in a metric space,” inAdvances in Neural Information Processing Systems (NeurIPS), 2017

    work page 2017

  12. [12]

    Tooth segmentation using generative adversarial networks,

    J.-H. Hwang, D.-Y . Kim, and B.-S. Shin, “Tooth segmentation using generative adversarial networks,”Pattern Recognition Letters, vol. 145, pp. 39–45, 2021

    work page 2021

  13. [13]

    TFormer: 3D tooth segmentation in mesh scans with geometry guided transformer,

    H. Xiong, K. Li, K. Tan, Y . Feng, J. T. Zhou, J. Hao, and Z. Liu, “TFormer: 3D tooth segmentation in mesh scans with geometry guided transformer,”arXiv preprint arXiv:2210.16627, 2022. [Online]. Available: https://arxiv.org/abs/2210.16627

    work page arXiv 2022

  14. [14]

    TeethNet: Dual-stream attention network for 3D tooth segmentation,

    Z. Cui, C. Li, N. Chen, G. Wei, R. Run, X. Liu, and D. Ren, “TeethNet: Dual-stream attention network for 3D tooth segmentation,”Computers in Biology and Medicine, vol. 136, p. 104670, 2021

    work page 2021

  15. [15]

    Robust segmentation of partial and imperfect dental arches,

    A. A. Alsheghri, Y . Zhang, G. Hosseinimanesh, J. Keren, F. Cheriet, and F. Guibault, “Robust segmentation of partial and imperfect dental arches,”Applied Sciences, vol. 14, no. 23, p. 10784, 2024. [Online]. Available: https://www.mdpi.com/2076-3417/14/23/10784

    work page 2024

  16. [16]

    CAD/CAM in dentistry — a critical review,

    A. P. B. Samra, E. C. C. de Morais, R. F. Mazur, S. R. Vieira, and R. N. Rached, “CAD/CAM in dentistry — a critical review,”Revista Odonto Ciˆencia, vol. 31, no. 3, pp. 140–144, 2016

    work page 2016

  17. [17]

    DCrownFormer: Morphology-aware point-to- mesh generation transformer for dental crown prosthesis from 3D scan data of antagonist and preparation teeth,

    S. Yang, J. Han, S.-H. Lim, J.-Y . Yoo, S. Kim, D. Song, S. Kim, J.- M. Kim, and W.-J. Yi, “DCrownFormer: Morphology-aware point-to- mesh generation transformer for dental crown prosthesis from 3D scan data of antagonist and preparation teeth,” inMedical Image Computing and Computer Assisted Intervention – MICCAI 2024. Springer Nature Switzerland, 2024, p...

    work page 2024

  18. [18]

    From mesh completion to AI-designed crown,

    G. Hosseinimanesh, F. Ghadiri, F. Guibault, F. Cheriet, and J. Keren, “From mesh completion to AI-designed crown,”arXiv preprint arXiv:2501.04914, 2025

    work page arXiv 2025

  19. [19]

    Personalized dental crown design: Point- to-mesh generation,

    G. Hosseinimaneshet al., “Personalized dental crown design: Point- to-mesh generation,”Polytechnique Montr ´eal Publications, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/ S1361841524003645

    work page 2025

  20. [20]

    Morphology and fracture behavior of lithium disilicate dental crowns designed by human and knowledge-based AI,

    H. Dinget al., “Morphology and fracture behavior of lithium disilicate dental crowns designed by human and knowledge-based AI,”Journal of Dentistry, vol. 122, p. 104146, 2022

    work page 2022

  21. [21]

    IntelliDent: an AI-based online automated framework for dental crown generation,

    I. Chafi, G. Hosseinimaneshet al., “IntelliDent: an AI-based online automated framework for dental crown generation,”ResearchGate Pub- lication, 2025

    work page 2025

  22. [22]

    Automated tooth crown design with optimized shape and biomechanics properties,

    S. Tianet al., “Automated tooth crown design with optimized shape and biomechanics properties,”Frontiers in Bioengineering and Biotechnol- ogy, 2023

    work page 2023

  23. [23]

    Model adaptive tooth segmentation,

    R. Chen, J. Yang, Y . Feng, J. Hao, and Z. Liu, “Model adaptive tooth segmentation,” inMedical Imaging with Deep Learning, ser. Proceedings of Machine Learning Research, I. Oguz, J. Noble, X. Li, M. Styner, C. Baumgartner, M. Rusu, T. Heinmann, D. Kontos, B. Landman, and B. Dawant, Eds., vol. 227. PMLR, 2024, pp. 775–798. [Online]. Available: https://proc...

    work page 2024

  24. [24]

    Unsupervised pre-training improves tooth segmentation in 3-dimensional intraoral mesh scans,

    X. He, H. Wang, H. Hu, J. Yang, Y . Feng, G. Wang, and Z. Liu, “Unsupervised pre-training improves tooth segmentation in 3-dimensional intraoral mesh scans,” inProceedings of The 5th International Conference on Medical Imaging with Deep Learning, ser. Proceedings of Machine Learning Research, E. Konukoglu, B. Menze, A. Venkataraman, C. Baumgartner, Q. Dou...

    work page 2022

  25. [25]

    A fast point clouds registration algorithm for laser scanners,

    G. Xu, Y . Pang, Z. Bai, Y . Wang, and Z. Lu, “A fast point clouds registration algorithm for laser scanners,”Applied Sciences, vol. 11, no. 8, p. 3426, 2021

    work page 2021

  26. [26]

    Automated monitoring of tooth wear progression using AI on intraoral scans,

    N. van Nistelrooij, E. Maier, H. Bronkhorst, L. Crins, T. Xi, B. A. C. Loomans, and S. Vinayahalingam, “Automated monitoring of tooth wear progression using AI on intraoral scans,”Journal of Dentistry, vol. 150, p. 105323, 2024. [Online]. Available: https://doi.org/10.1016/j.jdent.2024.105323

    work page doi:10.1016/j.jdent.2024.105323 2024

  27. [27]

    Teeth3DS: A benchmark for teeth segmentation and labeling from intra-oral 3D scans,

    A. Ben-Hamadou, N. Neifar, A. Rekik, O. Smaoui, F. Bouzguenda, S. Pujades, E. Boyer, and E. Ladroit, “Teeth3DS: A benchmark for teeth segmentation and labeling from intra-oral 3D scans,” arXiv preprint arXiv:2210.06094, 2022. [Online]. Available: https: //arxiv.org/abs/2210.06094

    work page arXiv 2022

  28. [28]

    Fast point feature histograms (FPFH) for 3D registration,

    R. B. Rusu, N. Blodow, and M. Beetz, “Fast point feature histograms (FPFH) for 3D registration,” inIEEE International Conference on Robotics and Automation (ICRA), 2009, pp. 3212–3217

    work page 2009

  29. [29]

    Object modelling by registration of multiple range images,

    Y . Chen and G. Medioni, “Object modelling by registration of multiple range images,” inImage and Vision Computing, vol. 10, no. 3, 1991, pp. 145–155

    work page 1991

  30. [30]

    Open3D: A Modern Library for 3D Data Processing

    Q.-Y . Zhou, J. Park, and V . Koltun, “Open3D: A modern library for 3D data processing,”arXiv preprint arXiv:1801.09847, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  31. [32]

    Available: https://arxiv.org/abs/2603.04771

    [Online]. Available: https://arxiv.org/abs/2603.04771

    work page arXiv

  32. [33]

    Sailer, G

    I. Sailer, G. I. Benic, V . Fehmer, C. H. F. H¨ammerle, and S. M¨uhlemann, “Randomized controlled within-subject evaluation of digital and conven- tional workflows for the fabrication of lithium disilicate single crowns. Part II: CAD-CAM versus conventional laboratory procedures,”Journal of Prosthetic Dentistry, vol. 118, no. 1, pp. 43–48, 2017

    work page 2017