OrthoAI: A Neurosymbolic Framework for Evidence-Grounded Biomechanical Reasoning in Clear Aligner Orthodontics

Edouard Lansiaux; Margaux Leman; Mehdi Ammi

cs.CVFeb 23, 2026

OrthoAI: A Neurosymbolic Framework for Evidence-Grounded Biomechanical Reasoning in Clear Aligner Orthodontics

Authors: Edouard Lansiaux, Margaux Leman, Mehdi Ammi

Organizations: STaR-AI, Emergency Department, Lille University Hospital, 59000, Lille, France · Artificial Intelligence and Data Semantics Laboratory, Paris 8 University, 93200, Saint-Denis, France

Abstract

Automated clinical decision support for clear aligner orthodontics faces a key challenge: bridging geometric perception (3D tooth segmentation) with clinical reasoning (biomechanical feasibility). We address this with OrthOAI, introducing three methodological contributions. First, sparse-supervision segmentation: a landmark-to-point-cloud synthesis protocol enables training from sparse anatomical annotations (6-8 points per tooth) instead of dense labels, combined with a clinically stratified loss mixing label-smoothed cross-entropy and a batch-adaptive Dice term for class imbalance. Second, knowledge-grounded constraint inference: biomechanical feasibility is modeled as a Constraint Satisfaction Problem over a domain ontology of tooth movements, encoding evidence-based per-stage limits as soft and hard constraints. Third, multi-criteria treatment evaluation: treatment quality is scored through a formal Multi-Criteria Decision Analysis framework using a weighted Additive Value Function grounded in clinical priority theory. On landmark-reconstructed point clouds from 3DTeethLand (MICCAI 2024), segmentation reaches 81.4% Tooth Identification Rate with 60,705 parameters. Ablations quantify the impact of each design choice. End-to-end inference runs in under 4 seconds on CPU. We also outline the gap between the current prototype-trained on synthetic ellipsoidal approximations-and clinical deployment, with a roadmap for validation. Code and weights are released.

Figures & tables

Movement	Tooth class	Limit	Type	Source
Bodily translation ( $t_{x},t_{y}$ )	All	0.25 mm	Soft	[ 11 ]
Intrusion ( $t_{z}<0$ )	All	0.25 mm	Soft	[ 2 ]
Extrusion ( $t_{z}>0$ )	Posterior	0.20 mm	Hard	[ 1 ]
Extrusion ( $t_{z}>0$ )	Anterior	0.15 mm	Hard	[ 1 ]
Mesiodistal tip ( $r_{y}$ )	All	0.25 mm eq.	Soft	[ 11 ]
Axial rotation ( $r_{z}$ )	Canine	2.0°	Soft	[ 2 ]

Table 1: Orthodontic CSP constraint specification. Domain limits

\bar{v}_{k,j}

by movement type and tooth class. Hard violation threshold

\alpha=1.5

Panel of Figure 1 — Figure 1: DGCNN_Seg architecture with receptive field interpretation. Each EdgeConv layer targets a different geometric scale (annotations below). Global max-pooling resolves arch-level symmetry ambiguities. Total: 60,705 parameters.

Model	mIoU	tIoU	Acc	TIR
DGCNN_Seg (ours, full)	0.0825	0.0617	0.345	0.814
– batch-adaptive Dice	0.0612	0.0481	0.302	0.743
– label smoothing	0.0781	0.0590	0.338	0.802
– global pooling	0.0644	0.0503	0.297	0.671
– multi-scale concat	0.0710	0.0541	0.318	0.758
PointNet (baseline)	0.0423	0.0311	0.228	0.612

Table 2: Segmentation results on the 3DTeethLand validation set (20 scans).

$k$	mIoU	TIR	Acc	Train loss
3	0.0825	0.814	0.345	2.152
5	0.0793	0.801	0.331	2.193
10	0.0718	0.776	0.312	2.267
20	0.0641	0.744	0.288	2.401

Table 3: Ablation over EdgeConv neighborhood size

k

. Training on 80 scans; best epoch reported.

Feature set	mIoU	TIR	Acc
Full 6-dim (ours)	0.0825	0.814	0.345
– arch-centroid dist ( $\\|\tilde{p}\\|$ )	0.0764	0.746	0.319
– height ( $\tilde{p}_{z}$ , duplicate)	0.0812	0.808	0.341
– radial dist ( $\sqrt{\tilde{p}_{x}^{2}+\tilde{p}_{y}^{2}}$ )	0.0791	0.783	0.332
XYZ only (3-dim)	0.0638	0.699	0.278

Table 4: Feature engineering ablation. Each row removes one dimension from the 6-dim feature vector.

Loss	mIoU	TIR	Acc	Convergence
CE only	0.0631	0.724	0.301	Stable
CE + FD	0.0711	0.769	0.328	Unstable (rare classes)
CE + BD (no LS)	0.0781	0.802	0.338	Stable
CE (LS) + BD (ours)	0.0825	0.814	0.345	Stable

Table 5: Loss function ablation. CE: cross-entropy; LS: label smoothing; BD: batch-adaptive Dice; FD: full-batch Dice.

Panel of Figure 2 — Figure 2: Validation metrics over 10 epochs. TIR and mIoU exhibit divergent optimization trajectories, empirically validating the perception-reasoning decoupling hypothesis (Section 3.2 ).

Panel of Figure 3 — Figure 3: Training loss curve over 10 epochs. Monotonic decrease from 3.17 to 2.15 confirms stable convergence.

Patient	Teeth	WAVF Score	Grade	Critical	Warning	Est. Duration
01WU1B7C	16	79	B	2	3	12 mo
13GRE9RZ	14	82	B	1	2	10 mo
3JIYO434	14	77	B	2	2	11 mo
233LFZAG	14	84	B	1	1	0 9 mo
Mean	14.5	80.5	B	1.5	2.0	10.5 mo

Table 6: Biomechanical CSP inference results. Index scale 0–100; grade thresholds: A

\geq

90, B

\geq

75, C

\geq

60.

Perturbed criterion	Max $\Delta S$	Relative $\Delta S$ / $\bar{S}$
Biomechanical compliance ( $w=0.30$ )	$\pm 7.2$ pts	$\pm 9.0\%$
Predictability ( $w=0.20$ )	$\pm 4.8$ pts	$\pm 6.0\%$
Staging efficiency ( $w=0.15$ )	$\pm 3.1$ pts	$\pm 3.9\%$
Attachment planning ( $w=0.15$ )	$\pm 2.9$ pts	$\pm 3.6\%$
IPR indication ( $w=0.10$ )	$\pm 1.8$ pts	$\pm 2.2\%$
Arch symmetry ( $w=0.10$ )	$\pm 1.6$ pts	$\pm 2.0\%$

Table 7: WAVF sensitivity analysis. Score variation (

\Delta S

) under

\pm 50\%

perturbation of individual weights.

System	Params	Supervision	mIoU †	CDSS	Formalism
PointNet++ [ 5 ]	1.0M	Full mesh	$\sim$ 85%	No	None
TSGCNet [ 9 ]	1.5M	Full mesh	$\sim$ 90%	No	None
MeshSegNet [ 6 ]	0.8M	Full mesh	$\sim$ 88%	No	None
ClinCheck (commercial)	N/A	N/A	N/A	Partial	None
OrthoAI (ours)	60.7K	LM only	8.25%	Yes (CSP+MCDA)	Full

Table 8: System-level comparison. “CDSS” indicates clinical decision support capability; “LM” landmark-only supervision; “Full” dense mesh annotations.

\dagger

: not directly comparable (different data and evaluation protocols).

3D Point Cloud Segmentation

3D point cloud segmentation assigns semantic, instance, or part labels to points for scene understanding, mapping, robotics, and scientific measurement.

See all 70 papers

Mar 10, 2026

OrthoAI v2: From Single-Agent Segmentation to Dual-Agent Treatment Planning for Clear Aligners

We present OrthoAI v2, the second iteration of our open-source pipeline for AI-assisted orthodontic treatment planning with clear aligners, substantially extending the single-agent framework previously introduced. The first version established a proof-of-concept based on Dynamic Graph Convolutional Neural Networks (\dgcnn{}) for tooth segmentation but was limited to per-tooth centroid extraction, lacked landmark-level precision, and produced a scalar quality score without staging simulation. \vtwo{} addresses all three limitations through three principal contributions: (i)~~a second agent adopting the Conditioned Heatmap Regression Methodology (\charm{})~~\cite{rodriguez2025charm} for direct, segmentation-free dental landmark detection, fused with Agent~1 via a confidence-weighted orchestrator in three modes (parallel, sequential, single-agent); (ii)~a composite six-category biomechanical scoring model (biomechanics

\times

0.30 + staging

\times

0.20 + attachments

\times

0.15 + IPR

\times

0.10 + occlusion

\times

0.10 + predictability

\times

0.15) replacing the binary pass/fail check of v1; (iii)~a multi-frame treatment simulator generating

F = A \times r

temporally coherent 6-DoF tooth trajectories via SLERP interpolation and evidence-based staging rules, enabling ClinCheck 4D visualisation. On a synthetic benchmark of 200 crowding scenarios, the parallel ensemble of OrthoAI v2 reaches a planning quality score of

92.8 \pm 4.1

vs.\

76.4 \pm 8.3

for OrthoAI v1, a

+21\%

relative gain, while maintaining full CPU deployability (

4.2 \pm 0.8

~s).

Lansiaux Edouard, Leman Margaux

STaR-AI Research Group, Emergency Department, Lille University Hospital

Topic pending

Ask PDF

Apr 1, 2026

MATHENA: Mamba-based Architectural Tooth Hierarchical Estimator and Holistic Evaluation Network for Anatomy

Dental diagnosis from Orthopantomograms (OPGs) requires coordination of tooth detection, caries segmentation (CarSeg), anomaly detection (AD), and dental developmental staging (DDS). We propose Mamba-based Architectural Tooth Hierarchical Estimator and Holistic Evaluation Network for Anatomy (MATHENA), a unified framework leveraging Mamba's linear-complexity State Space Models (SSM) to address all four tasks. MATHENA integrates MATHE, a multi-resolution SSM-driven detector with four-directional Vision State Space (VSS) blocks for O(N) global context modeling, generating per-tooth crops. These crops are processed by HENA, a lightweight Mamba-UNet with a triple-head architecture and Global Context State Token (GCST). In the triple-head architecture, CarSeg is first trained as an upstream task to establish shared representations, which are then frozen and reused for downstream AD fine-tuning and DDS classification via linear probing, enabling stable, efficient learning. We also curate PARTHENON, a benchmark comprising 15,062 annotated instances from ten datasets. MATHENA achieves 93.78% mAP@50 in tooth detection, 90.11% Dice for CarSeg, 88.35% for AD, and 72.40% ACC for DDS.

Kyeonghun Kim, Jaehyung Park, Youngung Han +17

OUTTA, Seoul, South Korea · Seoul National University, Seoul, South Korea · Sangmyung University, Seoul, South Korea +4

Dental Imaging Medical Imaging Medical Image Segmentation Medical Image Anomaly Detection Object Detection State Space Models

Ask PDF

Mar 20, 2026

TSegAgent: Zero-Shot Tooth Segmentation via Geometry-Aware Vision-Language Agents

Automatic tooth segmentation and identification from intra-oral scanned 3D models are fundamental problems in digital dentistry, yet most existing approaches rely on task-specific 3D neural networks trained with densely annotated datasets, resulting in high annotation cost and limited generalization to scans from unseen sources. Thus, we propose TSegAgent, which addresses these challenges by reformulating dental analysis as a zero-shot geometric reasoning problem rather than a purely data-driven recognition task. The key idea is to combine the representational capacity of general-purpose foundation models with explicit geometric inductive biases derived from dental anatomy. Instead of learning dental-specific features, the proposed framework leverages multi-view visual abstraction and geometry-grounded reasoning to infer tooth instances and identities without task-specific training. By explicitly encoding structural constraints such as dental arch organization and volumetric relationships, the method reduces uncertainty in ambiguous cases and mitigates overfitting to particular shape distributions. Experimental results demonstrate that this reasoning-oriented formulation enables accurate and reliable tooth segmentation and identification with low computational and annotation cost, while exhibiting strong generalization across diverse and previously unseen dental scans.

Shaojie Zhuang, Lu Yin, Guangshun Wei +3

Shandong University, Jinan 250010, China · University of Surrey, Guildford, Surrey, United Kingdom · King’s College London, London, United Kingdom

Medical Imaging Dental Imaging Medical Image Segmentation 3D Instance Segmentation 3D VLMs

Ask PDF

Abstract

3D Point Cloud Segmentation

Explore similar work

OrthoAI v2: From Single-Agent Segmentation to Dual-Agent Treatment Planning for Clear Aligners

MATHENA: Mamba-based Architectural Tooth Hierarchical Estimator and Holistic Evaluation Network for Anatomy

TSegAgent: Zero-Shot Tooth Segmentation via Geometry-Aware Vision-Language Agents