TArchVGAN: Automatic Tooth Arrangement via Dental Arch Variability-Aware Gan

Accurate tooth arrangement is crucial for predicting optimal movement in digital orthodontics. Existing deep learning methods often overlook the inherent variability of dental arches, particularly the spatial changes in tooth centroids before and after arrangement. This variability is vital for evaluating tooth alignment outcomes. To address this, we propose TArchVGAN, a novel approach that integrates adversarial training for precise tooth arrangement. TArchVGAN comprises an automatic tooth arrangement network and a dental arch variability-aware discriminator. The network optimizes tooth positions by predicting transformation matrices from point clouds of misaligned teeth, while the discriminator evaluates the authenticity of predicted arches against the true target arch, based on the original misaligned arch. This adversarial training allows the network to learn arch variability and incorporate it into the optimization process. Experimental results demonstrate that TArchVGAN significantly outperforms existing techniques, highlighting its substantial potential for practical applications in digital orthodontics.