人工智能在颞下颌关节影像中的应用研究进展

Research Progress on the Application of Artificial Intelligence in Temporomandibular Joint Imaging

摘要: 人工智能(AI)技术，特别是深度学习算法，在颞下颌关节(TMJ)影像诊断中的应用正迅速发展，显著提升了诊断效率与准确性。本文系统综述了AI在TMJ不同成像模态中的研究进展，涵盖锥形束CT (CBCT)、磁共振成像(MRI)、曲面体层片、头颅侧位片及超声、红外热成像等多模态影像。在CBCT方面，卷积神经网络(CNN)被广泛用于髁突与关节窝的自动分割与骨关节炎(TMJOA)分类，分割精度可达Dice系数0.98，诊断准确率超过90%。MRI方面，AI模型在关节盘分割与移位诊断中表现优异，AUC值最高达0.99，且可识别关节积液、穿孔及退行性变。AI在曲面体层片与头颅侧位片中也实现了TMJOA的自动筛查与分类，准确率接近专家水平。此外，生成对抗网络(GAN)被用于MRI图像合成，增强数据多样性；多模态融合技术则提升了动态运动分析的精度。尽管AI在TMJ影像中展现出巨大潜力，但仍面临数据标准化、模型泛化及伦理法规等挑战。未来，随着算法优化与临床验证的深入，AI有望实现TMJ疾病的精准、高效与个性化诊疗。

Abstract: The application of artificial intelligence (AI) technology, especially deep learning algorithms, in the imaging diagnosis of temporomandibular joint (TMJ) is developing rapidly, significantly enhancing diagnostic efficiency and accuracy. This article systematically reviews the research progress of AI in different imaging modalities of TMJ, including cone-beam computed tomography (CBCT), magnetic resonance imaging (MRI), panoramic radiographs, lateral cephalograms, and ultrasound, infrared thermography, etc. In CBCT, convolutional neural networks (CNN) are widely used for the automatic segmentation of condyles and glenoids and the classification of temporomandibular joint osteoarthritis (TMJOA), with segmentation accuracy reaching a Dice coefficient of 0.98 and diagnostic accuracy exceeding 90%. In MRI, AI models perform excellently in the segmentation of articular discs and the diagnosis of disc displacement, with the highest AUC value reaching 0.99, and can also identify joint effusion, perforation, and degeneration. AI has also achieved automatic screening and classification of TMJOA in panoramic radiographs and lateral cephalograms, with accuracy approaching that of experts. In addition, generative adversarial networks (GAN) are used for MRI image synthesis to enhance data diversity; multimodal fusion technology improves the accuracy of dynamic motion analysis. Although AI shows great potential in TMJ imaging, it still faces challenges such as data standardization, model generalization, and ethical regulations. In the future, with the deepening of algorithm optimization and clinical validation, AI is expected to achieve precise, efficient, and personalized diagnosis and treatment of TMJ diseases.