Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OrthoAI: A Lightweight Deep Learning Framework for Automated Biomechanical Analysis in Clear Aligner Orthodontics -- A Methodological Proof-of-Concept

Edouard Lansiaux, Margaux Leman, Mehdi Ammi

TL;DR

OrthoAI, an open-source proof-of-concept decision-support system combining lightweight 3D dental segmentation with automated biomechanical analysis to assist treatment-plan evaluation, is presented.

Abstract

Clear aligner therapy now dominates orthodontics, yet clinician review of digitally planned tooth movements-typically via ClinCheck (Align Technology)-remains slow and error-prone. We present OrthoAI, an open-source proof-of-concept decision-support system combining lightweight 3D dental segmentation with automated biomechanical analysis to assist treatment-plan evaluation. The framework uses a Dynamic Graph CNN trained on landmark-reconstructed point clouds from 3DTeethLand (MICCAI) and integrates a rule-based biomechanical engine grounded in orthodontic evidence (Kravitz et al 2009; Simon et al 2014). The system decomposes per-tooth motion across six degrees of freedom, computes movement-specific predictability, issues alerts when biomechanical limits are exceeded, and derives an exploratory composite index. With 60,705 trainable parameters, segmentation reaches a Tooth Identification Rate of $81.4\%$ and mIoU of $8.25\%$ on surrogate point clouds-reflecting sparse landmark supervision rather than dense meshes. Although spatial boundaries are coarse, downstream analysis depends mainly on tooth identity and approximate centroid/axis estimation. Results establish a baseline for future full-mesh training and highlight current perceptual limits. The end-to-end pipeline runs in $<4s$ on consumer hardware. Code, weights, and analysis tools are released to support reproducible research in geometric deep learning and digital orthodontics. The system has not been validated on real intraoral meshes and should not be assumed to generalize beyond landmark-derived representations.

OrthoAI: A Lightweight Deep Learning Framework for Automated Biomechanical Analysis in Clear Aligner Orthodontics -- A Methodological Proof-of-Concept

TL;DR

OrthoAI, an open-source proof-of-concept decision-support system combining lightweight 3D dental segmentation with automated biomechanical analysis to assist treatment-plan evaluation, is presented.

Abstract

Clear aligner therapy now dominates orthodontics, yet clinician review of digitally planned tooth movements-typically via ClinCheck (Align Technology)-remains slow and error-prone. We present OrthoAI, an open-source proof-of-concept decision-support system combining lightweight 3D dental segmentation with automated biomechanical analysis to assist treatment-plan evaluation. The framework uses a Dynamic Graph CNN trained on landmark-reconstructed point clouds from 3DTeethLand (MICCAI) and integrates a rule-based biomechanical engine grounded in orthodontic evidence (Kravitz et al 2009; Simon et al 2014). The system decomposes per-tooth motion across six degrees of freedom, computes movement-specific predictability, issues alerts when biomechanical limits are exceeded, and derives an exploratory composite index. With 60,705 trainable parameters, segmentation reaches a Tooth Identification Rate of and mIoU of on surrogate point clouds-reflecting sparse landmark supervision rather than dense meshes. Although spatial boundaries are coarse, downstream analysis depends mainly on tooth identity and approximate centroid/axis estimation. Results establish a baseline for future full-mesh training and highlight current perceptual limits. The end-to-end pipeline runs in on consumer hardware. Code, weights, and analysis tools are released to support reproducible research in geometric deep learning and digital orthodontics. The system has not been validated on real intraoral meshes and should not be assumed to generalize beyond landmark-derived representations.
Paper Structure (32 sections, 4 equations, 3 figures, 5 tables)

This paper contains 32 sections, 4 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. 3D Dental Segmentation
  4. Biomechanical Analysis in Clear Aligner Therapy
  5. Clinical Decision Support in Digital Orthodontics
  6. Materials and Methods
  7. Dataset: 3DTeethLand MICCAI 2024
  8. Point Cloud Generation from Landmarks
  9. Data Augmentation
  10. Network Architecture: DGCNN_Seg
  11. EdgeConv Layers
  12. Multi-Scale Aggregation and Global Pooling
  13. Segmentation Head
  14. Loss Function
  15. Training Protocol
  16. ...and 17 more sections

Figures (3)

  • Figure 1: DGCNN_Seg architecture. Four EdgeConv layers with dynamic $k$-NN ($k=3$) produce multi-scale features that are concatenated, fused, enriched with a global descriptor, and classified by a per-point MLP head. Total: 60,705 parameters.
  • Figure 2: Training loss curve (CE + Dice) over 10 epochs. The loss decreases monotonically from 3.17 to 2.15, indicating stable convergence without overfitting despite the small dataset.
  • Figure 3: Validation metrics across training epochs. TIR consistently exceeds 60%, peaking at 84.1% (epoch 7), while mIoU and accuracy fluctuate, reflecting the challenge of fine-grained boundary delineation from sparse landmark supervision.