IADR Abstract Archives
Dental Crown Design Using Artificial Intelligence (AI): Is It Feasible?
Objectives: Dental crown design is one of the most important aspects in restorative dentistry. It is not only the requirements for functional purposes such as mastication and phonetics, but also for aesthetic and confidence of the patient. The design must be personalised to fit individual’s conditions and requirements. Computer-aided design and computer-aided manufacturing (CAD/CAM) has been advocated to improve traditional workflow. However, it encounters problems such as lack of accuracy, expensive, and still need experienced practitioners to operate. A fully automatic algorithm for dental crown design by utilising AI technology is presented in this study with the potential of improving current workflow.
Methods: 500 sets of mandibular second premolars, their adjacent and antagonist teeth were collected digitally, and machine learned with Generative Adversarial Network (GAN) approach. 12 sets of data were randomly selected as test dataset. The 12 natural teeth in the test dataset were compared with (1) GAN design, (2) Biogeneric design by CEREC, and (3) technician’s design individually in parameters of 3D similarity, cusp angle, occlusal contact point number and area, and Finite Element (FE) static and fatigue simulation.
Results: GAN design and natural tooth had lowest discrepancy in morphology compared with other groups. Biogeneric design showed a significant (α=0.05) higher cusp angle compared with GAN design and natural tooth. No significant difference was observed regarding the occlusal contact point number and area among all four groups. FE analysis results showed GAN design had a comparable performance with natural teeth regarding the stress distribution in crown, adhesive layer and dentine; the two groups also showed similar fatigue lifetimes under simulated cyclic loadings of 100-400 N.
Conclusions: Dental crowns designed by the GAN method in this study showed no statistical differences among morphological, occlusal and mechanical parameters, i.e., AI designed crowns are comparable with natural teeth. This study demonstrated AI can be utilised to design personalised dental crowns with high accuracy.
Methods: 500 sets of mandibular second premolars, their adjacent and antagonist teeth were collected digitally, and machine learned with Generative Adversarial Network (GAN) approach. 12 sets of data were randomly selected as test dataset. The 12 natural teeth in the test dataset were compared with (1) GAN design, (2) Biogeneric design by CEREC, and (3) technician’s design individually in parameters of 3D similarity, cusp angle, occlusal contact point number and area, and Finite Element (FE) static and fatigue simulation.
Results: GAN design and natural tooth had lowest discrepancy in morphology compared with other groups. Biogeneric design showed a significant (α=0.05) higher cusp angle compared with GAN design and natural tooth. No significant difference was observed regarding the occlusal contact point number and area among all four groups. FE analysis results showed GAN design had a comparable performance with natural teeth regarding the stress distribution in crown, adhesive layer and dentine; the two groups also showed similar fatigue lifetimes under simulated cyclic loadings of 100-400 N.
Conclusions: Dental crowns designed by the GAN method in this study showed no statistical differences among morphological, occlusal and mechanical parameters, i.e., AI designed crowns are comparable with natural teeth. This study demonstrated AI can be utilised to design personalised dental crowns with high accuracy.