深度学习在正畸诊断和治疗中的应用进展

Advances in the Application of Deep Learning for Orthodontic Diagnosis and Treatment

摘要: 深度学习作为人工智能的重要分支，通过模拟人脑神经元工作机制处理复杂任务，近年来已成为口腔正畸领域精准医疗的关键技术。本文系统评述深度学习在正畸诊疗中的研究进展：在诊断环节，基于头影测量片、全景片及锥形束计算机断层扫描(CBCT)的深度学习技术，可实现颅面标志点自动识别、牙齿分割编号、三维结构分割及发育异常检测，显著提升分析效率并降低人工误差；在治疗规划中，该技术通过预测牙齿移动轨迹、辅助拔牙决策、生成个性化矫治方案及模拟正颌手术过程，为临床决策提供重要支持。当前临床应用仍面临数据来源单一、标注标准差异、模型可解释性不足及多中心泛化能力有限等挑战。未来需构建标准化多中心数据集、开发可解释性模型并优化临床系统集成，推动技术深度融入诊疗全流程。本综述介绍了深度学习技术在口腔正畸领域的应用现状，以增进读者了解。

Abstract: Deep learning, a vital branch of artificial intelligence, processes complex tasks by simulating the working mechanisms of neurons in the human brain. In recent years, it has become a key technology for precision medicine in orthodontics. This review systematically examines research advances in deep learning for orthodontic diagnosis and treatment. In diagnosis, deep learning techniques based on cephalometric radiographs, panoramic radiographs, and cone-beam computed tomography (CBCT) enable automated identification of craniofacial landmarks, tooth segmentation and numbering, 3D structural segmentation, and developmental anomaly detection, significantly improving analysis efficiency while reducing manual errors. For treatment planning, these techniques provide crucial decision support by predicting tooth movement trajectories, assisting in extraction decisions, generating personalized treatment plans, and simulating orthognathic surgical procedures. Current clinical applications still face challenges including limited data sources, inconsistent annotation standards, insufficient model interpretability, and restricted multi-center generalization capability. Future efforts should focus on establishing standardized multi-center datasets, developing interpretable models, and optimizing clinical system integration to facilitate deeper incorporation of the technology into the entire diagnostic and therapeutic workflow. This review presents the current applications of deep learning in orthodontics to enhance reader comprehension.