کنفرانس بین المللی فناوری اطلاعات و دانش

صفحه اصلی / شانزدهمین کنفرانس بین المللی فناوری اطلاعات و دانش

Multi-Modal Longitudinal Tooth Labeling with Temporal Graph–Transformer Integration

نویسندگان :

Maral Mirza mohammadi¹ Mahdi Tarom²

1- علم 2- مالک

کلمات کلیدی :

Tooth segmentation،longitudinal labeling،; graph neural networks; temporal refinement.

چکیده :

Abstract— Accurate and temporally consistent tooth labeling is essential for longitudinal dental analysis and treatment planning. This paper presents a unified framework for longitudinal 3D tooth labeling that integrates cone-beam computed tomography (CBCT) and intraoral scans (IOS) via a hybrid temporal Graph Neural Network–3D Transformer architecture. The proposed pipeline performs segmentation, tooth ID assignment, and cross-timepoint label propagation, followed by a temporal refinement module that enforces label consistency while preserving spatial accuracy. Evaluations on two public multi-timepoint datasets—STS-3D-Tooth and Teeth3DS+—achieved state-of-the-art performance with a Dice score of 95.5%, IoU of 91.0%, ID accuracy of 98.3%, and temporal label consistency (TLC) of 93.4%. Qualitative results further confirm robust handling of missing and erupting teeth with stable identity tracking across sessions. The proposed multimodal hybrid framework provides a clinically reliable and computationally efficient solution for automated longitudinal dental imaging.