Towards automated model analysis: A multiview AI segmentation of 3D dental scans

Precise identification of tooth and gingival boundaries in digital models is essential for effective orthodontic diagnosis, treatment planning, and appliance fabrication. Recent advances in artificial intelligence (AI) offer opportunities to automate this critical step with accuracy and generalizability. The objective of this study is to develop and evaluate an AI-based multiview segmentation approach that reduces manual workload, and supports daily orthodontic workflows using 3D intraoral scans.

A total of 1200 dental models from the public 3D Teeth Challenge dataset and 29 clinical 3D intraoral scans were used. Each 3D model was converted into multiple 2D images using a modified multiview approach. An AI segmentation model, based on the Mask2Former architecture, was trained to segment tooth boundaries automatically. Performance was assessed using mean Intersection over Union (mIoU) and Dice Similarity Coefficient (DICE) scores, comparing results to existing approaches.

The proposed model achieved high accuracy on both public and clinical datasets. On the public testing dataset, it reached an mIoU score of 93.1% ± 0.09 and a DICE score of 95.7% ± 0.09. On the clinical testing set, it maintained strong performance with an mIoU score of 90.7% ± 0.01 and a DICE score of 94.9% ± 0.01, demonstrating its ability to generalize to real-world intraoral scans.

This study demonstrates the clinical potential of our AI-based modified multiview segmentation model for intraoral scans. The approach provides accurate results across varying scan qualities, supporting efficient digital model analysis in orthodontics and promising routine clinical use.