Abstract: Traditional anthropometric measurements in clinical smile aesthetics practice are prone to subjective errors. This study proposes an intelligent prediction system for maxillary central incisor width, integrating automated 3D facial landmark detection, binocular 3D reconstruction, WGAN-GP data augmentation, and regression model analysis. Using 200 cases of 3D facial scan data, we established a feature space incorporating five key anatomical parameters including ICW and IAW. The innovative integration of Wasserstein Generative Adversarial Networks with gradient penalty mechanism effectively addressed model generalization challenges under small-sample conditions.A systematic comparison of five regression algorithms revealed that Gradient Boosting Regression (GBR) achieved a coefficient of determination (R2) of 0.9446 on the test set, with prediction error (RMSE=0.1238 mm) reduced by 73.4% compared to traditional methods, while Multilayer Perceptron (MLP) demonstrated superior generalization stability (test set R2=0.9691). Our method achieves submillimeter-level prediction accuracy (0.092-0.171 mm) through 3D feature space mapping and ensemble learning strategies. We recommend the clinical adoption of the MLP-GBR hybrid model architecture, which provides an intelligent decision-making model with strong interpretability and clinical adaptability for digital smile design.