Related Experiment Video
Updated: Jan 9, 2026
Digital Hybrid Model Preparation for Virtual Planning of Reconstructive Dentoalveolar Surgical Procedures
Published on: August 5, 2021
A deep learning system on monolithic implant-supported crown design: Evaluating AI-generated models against
Shuxin Ren1, Tao Wu2, Dong Zhou1
1Department of Oral Implantology, Peking University School and Hospital of Stomatology, Beijing, China.
Purpose:
To investigate the effectiveness of a novel transformer-based deep learning model for generating monolithic implant-supported crowns (ISCs), comparing its performance to traditional software-generated crowns.
Materials And Methods:
A total of 311 patients who received single implant placements in the posterior first molar area were included. Digital impressions as well as technician-designed crowns were imported into the AI model. The PoinTr architecture-based AI model was trained on 291 cases and validated using 20 additional cases. Four groups of ISC designs were evaluated: Group 1, Software1-generated crowns; Group 2, Software 2-generated crowns; Group 3, AI-generated crowns; and Group 4, technician-designed crowns. Accuracy was evaluated through overall contour deviation, occlusal morphology, proximal contact, and emergence profile.
Results:
Relative to the reference (Group 4), AI-generated crowns (Group 3) demonstrated significantly lower overall contour deviation (438.6 ± 169.4 µm) compared with both software-generated crowns (Groups 1-2: 679.5-705.1 µm, p < 0.01). Occlusal discrepancies were smaller and contact distributions more comparable to technician-designed crowns in the AI group, while software-generated crowns showed higher frequencies of premature contacts. For emergence profile, AI crowns predominantly reproduced concave contours and angles approximating technician-designed crowns, whereas software groups often produced convex forms. Across all automated groups (1-3), proximal contact adaptation remained less accurate than in technician-designed crowns.
Conclusions:
Within the limitations of this proof-of-concept study, AI-generated implant-supported crowns achieved closer approximation to technician designs than software-generated crowns, particularly in contour, occlusal morphology, and emergence profile.
