Abstract
The mandibular defect reconstruction algorithms that are based on deep learning demonstrate significant potential for efficiently addressing clinical challenges in oral and maxillofacial surgery. The algorithm heavily relies on high-quality datasets, which remain one of the main bottlenecks. In this study, the first clinically derived Mandibular Defect Dataset, comprising 147 models of various mandibular defects, is introduced. Each model is manually annotated by experienced surgeons, and all processing workflows strictly adhere to clinical standards. In contrast to previous datasets, this dataset accurately represents the complexity of clinical defect boundaries and the diverse anatomical structures of individual patients, making it a valuable resource for developing AI models that exhibit improved generalizability and adaptability in mandibular reconstruction. Additionally, the dataset provides HCL classification diagnoses and relevant information for each defect model, thereby supporting more diverse clinical research in the future.