Development and Validation of an AI-Driven Automated Tooth Segmentation Algorithm for Intraoral Scans

Rojin Adabdokht, Manuel Lagravere Vich, Wenji Cai, Victor de Miranda Ladewig, Hollis Lai

Keywords: Deep Learning, 3D Mesh Segmentation, Intraoral Scanning

TL;DR: We developed and validated an explainable AI model that accurately segments teeth from intraoral scans using a two-step deep learning pipeline, significantly improving efficiency and precision in digital dentistry.

Abstract: This study presents the development and validation of an AI-driven, explainable algorithm for automated tooth segmentation from intraoral scans. Manual segmentation is labor-intensive and prone to inconsistency, limiting its scalability in digital dentistry. To address this, we implemented a two-step deep learning pipeline incorporating YOLOv8 for object detection and U-Net for semantic segmentation, combined with a novel noise reduction strategy to enhance accuracy using smaller labeled datasets. The model was trained and validated on expert-annotated intraoral and cast scan datasets. It achieved a mean average precision (mAP@0.5) of 0.98 for detection and a Dice Similarity Coefficient (DSC) of 0.94 during internal validation. External validation on an independent dataset yielded a DSC of 0.83, with posterior teeth and lower arches segmented more accurately. These findings demonstrate the model’s potential to streamline clinical workflows and improve diagnostic consistency in digital dental applications.

Submission Number: 3