Deep learning (DL) has been applied to reconstruct missing tooth surfaces. Although promising, no current method ensures that DL-generated prosthesis simultaneously meet clinical requirements for accuracy, surface roughness, anatomical morphology, and mechanical properties across fabrication techniques. Furthermore, while both natural tooth and technician-designed prosthesis datasets are available, there has been no research on how to better use these two datasets. The purpose of this study is to address these issues. We developed a geometric processing method that combines modified Delaunay triangulation (DT) reconstruction to achieve accurate, mechanically suitable results from 256 × 256 depth maps. A Tooth Generative Adversarial Network (ToothGAN) was trained with specialized loss functions for anatomical features and smoothness using both natural and technician-designed datasets. The output was validated via 3D printing and in vitro testing. ToothGAN outperformed prior algorithms on natural tooth data across metrics including Root Mean Square Error (RMSE), Structural Similarity Index (SSIM), 3-Dimensional Route Mean Square Error (3DRMSE), and Visual Assessment (VA) score. The generated crowns met the mechanical standards such as roughness, and Sharp Mesh Corner Ratio (SMCR), making them suitable for precision 3-Dimensional manufacturing. Blending natural and technician-designed data improved learning of anatomical features like cusps and grooves, though some metrics such as groove distance and occlusal contact points were altered. ToothGAN satisfies precision manufacturing demands and shows strong potential for clinical application in crown generation. • By modifying DT reconstruction with digital geometric processing can generate accurate and mechanically-suitable dental crowns. • ToothGAN with a new loss function for anatomical tooth features and smooth surfaces outperforms prior algorithms with natural tooth datasets. • Clinically, ToothGAN-designed crowns were on par with technician-designed crowns but better uniformity in occlusal contact.

Tooth segmentation is an essential step in digital dental diagnosis and treatment planning, playing a crucial role across various dental specialties. Despite its importance, this process is fraught with challenges due to the high noise and low contrast inherent in 2D and 3D dental radiographic images. Both convolutional neural networks (CNNs) and transformers have shown promise in tooth image segmentation, yet each method has limitations in handling long-range dependencies and computational complexity. To address this issue, this paper introduces T-Mamba, integrating frequency-based features and shared bi-positional encoding into vision mamba to address limitations in modeling global features efficiently. Besides, we designed a gate selection unit to integrate two features in the spatial domain and one feature in the frequency domain adaptively. T-Mamba is the first to integrate frequency-based features into vision mamba, offering the flexibility to process both 2D and 3D tooth image data without the need for separate modules. Additionally, the TED, a large-scale public 2D dental X-ray dataset, has been presented in this paper. Extensive experiments demonstrated that T-Mamba achieved new SOTA performance on both the 3D dental CBCT dataset and our proposed 2D TED dataset. The code and models are publicly available athttps://github.com/isbrycee/T-Mamba.