Progress of Mandibular Canal Segmentation: From Traditional Methods to AI-Driven Clinical Applications

下颌管(MC)是口腔颌面部区域的重要解剖结构，其精准分割对于下颌第三磨牙拔除术、牙种植术中下牙槽神经保护具有重要的临床意义。然而，由于下颌管体积小、边界模糊、个体差异大，使得传统方法在CBCT影像上的分割精度有限。近年来，基于深度学习的方法显著提升了下颌管自动分割的准确性和泛化能力。本文系统综述了下颌管分割的研究进展，涵盖传统方法、基于三维卷积神经网络的方法、基于Transformer的方法，以及半监督与弱监督学习策略。并展望了该领域的未来应用潜力，以期为MC自动分割的进一步发展及临床转化提供参考。The Mandibular Canal (MC) is an important anatomical structure in oral and maxillofacial regions, and its accurate segmentation is critical for mandibular third molar extraction and the protection of the inferior alveolar nerve in dental implant surgery. However, due to its small volume, blurred boundaries, and significant anatomical variability among individuals, traditional methods show limited segmentation accuracy on CBCT images. Recently, deep learning-based methods have significantly improved the accuracy and generalizability of automated MC segmentation. This paper systematically reviews the progress in MC segmentation, covering traditional methods, approaches based on three-dimensional convolutional neural networks (3D CNNs), Transformer-based methods, and semi-supervised or weakly supervised learning strategies. Additionally, the potential future applications in this field are discussed, aiming to provide references for further development and clinical translation of automated MC segmentation.