Abstract: Objective　To propose a learning-based automatic restoration method for craniomaxillofacial  defects, which learning from a self-constructed dataset to automatically generate the shape of the defective parts, and providing reference for the restoration of complex craniomaxillofacial  structures.　Methods　Based on the head CT data, 125 cases of skull data were reconstructed and annotated. Each case was categorized into 21 defect classes. Various techniques were used for data preprocessing, including image registration and threshold filtering. A novel craniomaxillofacial  automatic restoration technique was introduced to generate shapes for the defective portions.　Results　The proposed method achieves the state-of-the-art results on the CMF Defects dataset, which can reconstruct shapes that combine aesthetics and protective functionality.  Conclusions　Craniomaxillofacial bone have diverse shapes and complex anatomical structures. This study, combined with deep learning and data-driven methods, can effectively generate the generation of craniomaxillofacial bone defects, providing a reliable foundation for preoperative planning and intraoperative procedures in craniomaxillofacial restoration surgery.

Key words: Craniomaxillofacial restoration; ,  Image processing; ,  3D reconstruction;　Deep learning