Abstract

Deep learning has gained popularity in oral medicine for dental diagnostics, but traditional approaches heavily rely on clinical data that are often difficult to obtain. This study aims to integrate physical information from 3D elastic mechanics into the analysis of tooth stress, enabling direct evaluation from Cone Beam Computed Tomography (CBCT) data. A self-developed algorithm was used to automatically construct 3D tooth geometries from CBCT scan data. Physical information of 3D elastic mechanics was incorporated into the loss function of deep neural networks. Network parameters are optimized using the full-batch gradient descent algorithm, and Physics-Informed Neural Networks (PINNs) with specific boundary conditions for 3D elastic models are employed to compute physical information, including displacements and von Mises stress. The proposed PINNs successfully enable direct evaluation of tooth displacements and von Mises stress from CBCT data by integrating 3D elastic mechanics into the deep learning framework. The proposed approach can reduce reliance on scarce clinical data, providing a promising tool for non-invasive and efficient tooth stress assessment with potential implications for improving dental diagnostic workflows.

IPC Classification

Keywords