Multimodal Contrastive Pretraining of CBCT and IOS for Enhanced Tooth Segmentation

Abstract

Digital dentistry represents a transformative shift in modern dental practice. The foundation of this transformation lies in the accurate digital representation of a patient’s dentition, obtained from segmented Cone-Beam Computed Tomography (CBCT) and Intraoral Scans (IOS). Despite increasing interest in digital dental technologies, existing segmentation methods often lack rigorous validation and demonstrate limited clinical applicability. This work presents ToothMCL, the first multimodal pretraining framework for dental segmentation, addressing a critical and previously unmet challenge. Unlike prior single-modality approaches, ToothMCL integrates volumetric (CBCT) and surface-based (IOS) imaging through contrastive learning to capture modality-invariant anatomical representations. Using a curated dataset of 3,867 paired CBCT–IOS samples, ToothMCL achieves state-of-the-art performance, improving Dice scores by 12% for CBCT and 8% for IOS across the largest and most diverse evaluation to date. Our findings highlight the transformative potential of large-scale multimodal pretraining to advance clinical workflows in digital dentistry.

Date of Award	2026
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Hao CHEN (Supervisor)