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

A research team from Université de Lille has released OrthoAI, an open-source proof-of-concept framework that applies lightweight deep learning to automate biomechanical analysis for clear aligner orthodontic treatment. The system addresses a critical bottleneck in modern orthodontics, where clinicians currently manually review digitally planned tooth movements via platforms like Align Technology's ClinCheck—a process that remains slow and prone to human error. OrthoAI combines 3D dental segmentation with rule-based biomechanical analysis to create a decision-support tool that can rapidly evaluate treatment plans, decomposing per-tooth motion across six degrees of freedom and computing movement-specific predictability.

The technical core uses a Dynamic Graph CNN trained on landmark-reconstructed point clouds from the 3DTeethLand dataset, integrating a biomechanical engine grounded in established orthodontic evidence. With only 60,705 trainable parameters, the model achieves an 81.4% Tooth Identification Rate and 8.25% mIoU on surrogate point clouds, reflecting its sparse landmark supervision approach rather than dense mesh training. While spatial boundaries remain coarse and the system hasn't been validated on real intraoral scans, it establishes a baseline for future research and demonstrates that downstream analysis depends primarily on tooth identity and approximate centroid/axis estimation. The complete pipeline runs in under 4 seconds on consumer hardware, and the team has released all code, weights, and analysis tools to support reproducible research in geometric deep learning for digital orthodontics.