A fully automated database-driven framework for interproximal tooth morphology reconstruction in orthodontics

Due to crown overlap and insufficient scanner resolution, 3D tooth model obtained from intraoral scans often exhibit inter-tooth adhesion, resulting in loss of individual tooth interproximal morphology and blurred interdental spaces, which severely compromises the accuracy and efficacy of orthodontic treatment. Existing reconstruction methods rely heavily on manual intervention, limiting their clinical efficiency. To address this, we propose a fully automated database-driven framework that reconstructs missing tooth morphology through parametric template retrieval and deformation. Our method first constructs a parametric tooth database using multi-view convolutional neural networks (MVCNN), encoding 3D morphology into discriminative feature descriptors. A coarse-to-fine localization strategy enables fully automated localization with sub-millimeter accuracy. Missing morphology are then restored via iterative Laplacian deformation with weight constraints, while parametric B-spline modeling reconstructs root anatomy. Validation on clinical cases, our method achieved a root mean square surface distance of less than 0.096 mm, outperforming state-of-the-art approaches. The results demonstrate that our framework enables efficient and precise fully automated tooth reconstruction, offering a clinically viable solution for digital orthodontics.