Research Article
3D Cycle-Consistent Adversarial Network for Designing Dental Implant Crown
@INPROCEEDINGS{10.1007/978-3-031-84312-9_13, author={Georgi Kostadinov and Aleksandar Naydenov}, title={3D Cycle-Consistent Adversarial Network for Designing Dental Implant Crown}, proceedings={Computer Science and Education in Computer Science. 20th EAI International Conference, CSECS 2024, Sofia, Bulgaria, June 28--30, 2024, Proceedings}, proceedings_a={CSECS}, year={2025}, month={3}, keywords={3D Generative Adversarial Network Dental Prostheses AI in Dentistry 3D CycleGAN Voxelization Restorative Dentistry Dental Crown Design Automated Dental Design Intersection over Union Hausdorff Distance}, doi={10.1007/978-3-031-84312-9_13} }- Georgi Kostadinov
Aleksandar Naydenov
Year: 2025
3D Cycle-Consistent Adversarial Network for Designing Dental Implant Crown
CSECS
Springer
DOI: 10.1007/978-3-031-84312-9_13
Abstract
In this paper, we present a novel approach for designing screw retained crown over dental implant using a 3D Generative Adversarial Network (GAN), specifically 3D CycleGAN. The adversarial network was trained and validated on 3D intraoral scans from 150 patients which were adjusted for the study with ExoCad and then voxelized to a resolution of 64 × 64 × 64. Our results show an average Intersection over Union (IoU) of 75% and a mean Hausdorff distance of 1.0555 mm. This suggests a strong correlation between the generated crowns and manually designed ones, ensuring both functional and aesthetic suitability. Additionally, we generated visualizations and Hausdorff distance heatmaps to assess the alignment and deviations of the generated prostheses. The proposed approach overcomes the limitations of existing methods by fully incorporating the specific morphology of natural dental crown in the prosthesis design, resulting in crowns that are anatomically and functionally suitable for practical applications but designed without human intervention. Future enhancements include expanding the dataset with a higher variability of dental structures and increasing the input resolution of the proposed 3D CycleGAN network. Overall, our findings highlight the potential of machine learning to significantly improve the quality and efficiency of dental prosthesis design.
