Assessment of Creativity Potential of a 3DGAN in Implant Crown Design: A Proof-of-Concept Study

Digital dentistry increasingly relies on artificial intelligence (AI) to automate restorative design. However, the ability of generative networks to produce multiple geometrically distinct outputs for the same prosthetic field remains insufficiently evaluated. This study assessed repeated-output geometric variability in a previously developed three-dimensional generative adversarial network (3DGAN) for screw-retained implant crown design as a preliminary indicator of potential generative diversity. Nine AI-generated implant crown designs were analyzed, consisting of three independently generated crowns for each of three different prosthetic fields. Within each set, the crowns were superimposed and compared using "MeshLab". Mean Hausdorff distance (HD), maximum HD, and root mean square (RMS) values were recorded, with 0.05 model units used as the threshold for identifying insufficient morphological variation. The overall mean HD was 3.32 model units, the mean maximum HD was 16.18 model units, and the mean RMS value was 4.40 model units. No pairwise comparison showed values equal to or below 0.05 model units. In conclusion, the investigated 3DGAN demonstrated preliminary evidence of geometric output variability compatible with potential generative diversity.