IADR Abstract Archives
Accuracy of Artificial Intelligence-Designed Single-Tooth Dental Prostheses: a Feasibility Study
Objectives: Dental prostheses aim to restore patients’ oral functions and esthetics through biomimetic. This study aimed to determine the accuracy of an Artificial Intelligence (AI) designed single-tooth dental prosthesis with respect to biomimetics.
Methods: Digitized dental models of subjects satisfying inclusion/exclusion criteria were obtained using a laboratory scanner - Trios D2000 (3Shape, Denmark). Duplicate digitized models were created omitting a single tooth (right-maxillary-first-molars; FDI tooth 16). AI-designed single-tooth dental prosthesis were produced through backpropagation within a Generative Adversarial Networks (GAN) on a training data set. Accuracy of the AI-designed single-tooth dental prosthesis to replace omitted tooth 16 with respect to its natural healthy form was determined by calculating the Hausdorff distances between the comparing 3D surface models (AI-generated prosthesis versus original tooth).
Results: One- hundred- and-fifty-nine digitized models were used for training and feasibility/ accuracy was tested on 10 models. The mean Hausdorff distance between the 3D surface of AI-generated single-tooth prosthesis and its original counterpart ranged from 0.441 mm to 1.039 mm, with average root-mean-square errors of 1.046. In over half the cases, the AI-generated tooth was able to match with the corresponding natural healthy tooth (i.e. lowest mean Hausdorff distance).
Conclusions: This study demonstrates the feasibility of AI-designed single-tooth dental prosthesis with promising accuracy in terms of biomimetic reconstruction. With saturated training and improved algorithm, the AI may generate designs for dental prosthesis with accuracy and precision. This study has implications for the use of AI in future clinical applications in dentistry and specifically prosthodontics.
Methods: Digitized dental models of subjects satisfying inclusion/exclusion criteria were obtained using a laboratory scanner - Trios D2000 (3Shape, Denmark). Duplicate digitized models were created omitting a single tooth (right-maxillary-first-molars; FDI tooth 16). AI-designed single-tooth dental prosthesis were produced through backpropagation within a Generative Adversarial Networks (GAN) on a training data set. Accuracy of the AI-designed single-tooth dental prosthesis to replace omitted tooth 16 with respect to its natural healthy form was determined by calculating the Hausdorff distances between the comparing 3D surface models (AI-generated prosthesis versus original tooth).
Results: One- hundred- and-fifty-nine digitized models were used for training and feasibility/ accuracy was tested on 10 models. The mean Hausdorff distance between the 3D surface of AI-generated single-tooth prosthesis and its original counterpart ranged from 0.441 mm to 1.039 mm, with average root-mean-square errors of 1.046. In over half the cases, the AI-generated tooth was able to match with the corresponding natural healthy tooth (i.e. lowest mean Hausdorff distance).
Conclusions: This study demonstrates the feasibility of AI-designed single-tooth dental prosthesis with promising accuracy in terms of biomimetic reconstruction. With saturated training and improved algorithm, the AI may generate designs for dental prosthesis with accuracy and precision. This study has implications for the use of AI in future clinical applications in dentistry and specifically prosthodontics.