論文ID: JPR_D_25_00041
Purpose: This study proposes the development of “Motion-DSD”, an artificial intelligence-assisted workflow for digital smile design (DSD), which enables a dynamic 2-dimensional (2D) simulation of digital diagnostic waxing by transferring an intraoral design onto a frontal facial video, and validates its clinical feasibility.
Methods: A total of 2,000 facial and 190 intraoral images were used to fine-tune the pre-trained neural network, Segment Anything model (SAM), via two sets of low-rank adaptation (LoRA) modules for facial structures and teeth segmentation respectively. A transformation algorithm incorporating a “standard” facial image was developed to align intraoral and facial structures. A Flask-based web user interface (web-UI) was developed for clinical deployment. A participant sample set was prepared to validate the workflow’s performance in a clinical setting.
Results: Two fine-tuned SAMs achieved robust segmentation performance, with a mean Dice score coefficient of 0.886 for the facial dataset and 0.969 for the intraoral dataset. The alignment algorithm effectively transferred the intraoral DSD design onto the participant’s frontal facial video and enabled a 2D simulation of digital diagnostic waxing under various facial expressions, demonstrating its clinical feasibility. The web-UI allows dentists to interactively refine the design and preview simulation results in real time.
Conclusions: Motion-DSD enables the 2D simulation of digital diagnostic waxing from intraoral DSD designs in a dynamic facial context. The workflow overcomes the limitations of static imaging methods and manual alignment, bringing dynamics prior to the physical mockup phase. Further investigations are warranted to quantitatively validate the simulation accuracy and demonstrate its potential advantages over conventional static methods.
