Self-Structure Enhance Network for Digital Molar Wax-up Design

Abstract

Advancements in AI have driven digital innovation in oral healthcare, yet manual intervention still plays a critical role in dental restoration design. Current learning-based approaches are primarily focused on generating simpler dental structures, such as crowns or incisors, rather than addressing the more complex morphology of molars. To address this issue, in this paper, we propose a self-structure enhance network for digital molar wax-up design. Our work directly learns from unlabeled dental data, identifying multimodal feature and topological relationships between teeth, crown surfaces and gum. First, a feature extraction block is designed specifically for dental models, learning the spatial significance of positional data and enhancing local detail by assigning weights to key points. Next, a self-structuring enhancement block improves the 3D structural representation by utilizing multi-view depth maps. Finally, a missing position-aware block predicts locally weighted offsets to enrich the predicted surface features of the tooth. Extensive experiments demonstrate that our work achieves state-of-the-art performance in molar prediction across public and private datasets.