AI Research for Orthodontics

The Intelligence Behind OrthoAssist

Deep learning models trained on real clinical data. Tooth movement prediction, automated segmentation, cephalometric analysis. Peer-reviewed research powering clinical decisions.

71.8%
Prediction Accuracy
662K
Model Parameters
44
Training Cases
9+
Movement Types
AI Models

Five Models. One Mission.

Each model addresses a specific clinical challenge in orthodontic treatment planning.

Tooth Movement Prediction

Multi-task neural network predicting required tooth movements from pre-treatment 3D scans. Predicts both movement type (rotation, intrusion, expansion) and magnitude.

71.8%
accuracy

Tooth Segmentation (MeshSegNet)

Published in IEEE TMI. Automatically labels individual teeth from intraoral scans. 14 teeth + gingiva segmentation from raw 3D meshes.

IEEE
published

Cephalometric Analysis

AI landmark detection on X-rays and facial photos. 19 anatomical landmarks, skeletal angles (SNA, SNB, ANB), and dental measurements.

19
landmarks

Case Similarity Search

Deep learning embeddings from 3D scans enable semantic similarity search. Find past cases with similar malocclusions for treatment reference.

44
cases

Treatment Outcome Prediction

Predict treatment duration, complexity scores, and expected outcomes. Data-driven insights from real clinical results.

ML
powered
How It Works

From Raw Data to Clinical Intelligence

Step 1

Data Collection

3D intraoral scans (STL), CBCT volumes, clinical photos, and treatment metadata from practicing orthodontists.

Step 2

Model Training

Multi-modal deep learning with data augmentation, cross-validation, and biological constraint enforcement.

Step 3

Clinical Deployment

Trained models are deployed in OrthoAssist AI Pro, providing real-time predictions and treatment guidance.

Research Foundation

Built on Published Science

Every prediction is grounded in peer-reviewed orthodontic research and validated against clinical outcomes.

MeshSegNet published in IEEE Transactions on Medical Imaging + MICCAI 2019

Movement limits based on Kravitz, Haouili, Simon et al. research

Predictability rates validated: 72% mesial-distal, 47% bucco-lingual, 30% extrusion

Multi-task learning: simultaneous classification and regression on tooth movements

Peer-Reviewed Publications

Kravitz et al. (2009)How well does Invisalign work? A prospective clinical study evaluating the efficacy of tooth movement with Invisalign

AJO-DO · PMID: 19154919

Haouili et al. (2020)Has Invisalign improved? A prospective follow-up study evaluating the efficacy of tooth movement with Invisalign

AJO-DO · PMID: 32241352

Simon et al. (2014)Treatment outcome and efficacy of an aligner technique regarding incisor torque, premolar derotation and molar distalization

BMC Oral Health · PMID: 24352707

Weltman et al. (2010)Root resorption associated with orthodontic tooth movement: A systematic review

AJO-DO · PMID: 20197161

Lian et al. (2020)Deep Multi-Scale Mesh Feature Learning for Automated Labeling of Raw Dental Surfaces

IEEE TMI, Vol. 39, No. 7 · DOI: 10.1109/TMI.2020.2971730

Lian et al. (2019)MeshSegNet: Deep Multi-scale Mesh Feature Learning for End-to-End Tooth Labeling on 3D Dental Surfaces

MICCAI 2019, LNCS vol 11769 · Springer

Feng et al. (2018)Wing Loss for Robust Facial Landmark Localisation with Convolutional Neural Networks

CVPR 2018

Proffit WR, Fields HW, Sarver DM (2018)Contemporary Orthodontics, 6th Edition

Elsevier · ISBN: 978-0-323-54387-3

Nelson SJ (2014)Wheeler's Dental Anatomy, Physiology, and Occlusion, 10th Edition

Elsevier · ISBN: 978-0-323-26323-8

Open Source Foundation

MeshSegNetMIT

Deep multi-scale mesh feature learning for automated tooth labeling

Chunfeng Lian et al.

PointNetMIT

Point cloud processing architecture for 3D deep learning

Charles R. Qi et al.

vedoMIT

VTK-based 3D visualization and mesh processing toolkit

Marco Musy

LlamaIndexMIT

RAG framework for clinical knowledge retrieval

Jerry Liu et al.

Scientific Methods & Algorithms

Marching CubesTaubin SmoothingPointNet ArchitectureMulti-Task LearningGraph-Constrained LearningWing LossHeatmap RegressionData Augmentation (20×)Cross-ValidationVector Embedding SearchRAG PipelineB-Spline Fitting
Collaborate

Interested in the Research?

We're looking for research partners, clinical data contributors, and institutions interested in advancing AI in orthodontics.

Get in Touch