Article ID: JPR_D_24_00112
Purpose: Artificial intelligence (AI) may be used to learn and predict the maxillomandibular relationship, particularly when the number of occluding teeth pairs is insufficient. This study aimed to investigate the feasibility of training a new two-stage coarse-to-fine teeth alignment pipeline AI system in predicting maxillomandibular relationships based on the occlusal morphology of antagonistic teeth.
Methods: Maxillary and mandibular stone casts were collected and scanned at the maximal intercuspal position (MIP). A deep learning alignment network was trained using 90% of cast pairs. The remaining 10% of pairs were input into the trained AI system for validation. The maxillomandibular relationships predicted by the AI system were superimposed and compared with those of the mounted casts. Cartesian x-, y-, and z-coordinates were defined for each mandibular tooth scan with respect to (w.r.t.) its occlusal plane and dental midline. The discrepancy in the position of maxillary teeth scans was described based on rotation and translation.
Results: A total of 325 pairs of maxillary and mandibular stone casts were collected, with 300 pairs used for training and 25 for validation. For the AI-predicted maxillomandibular relationship, the mean rotational discrepancies w.r.t. the x-, y-, and z-axis were 1.407°±1.548°, 1.269°±8.476°, and 0.730°±1.334°, respectively. The mean translational discrepancies w.r.t. the x-, y-, and z-axis were 0.185±1.324 mm, 1.222±0.848 mm, -1.034±0.273 mm, respectively.
Conclusions: The AI-predicted maxillomandibular relationship for maxillary and mandibular teeth scans shows discrepancies of less than 1.3 mm and 1.5° compared to the actual relationships.
