超声影像组学在甲状腺结节良恶性鉴别诊断中的研究进展

Research Progress of Ultrasound Radiomics in the Differential Diagnosis of Benign and Malignant Thyroid Nodules

摘要: 甲状腺结节检出率逐年上升，如何无创、精准地鉴别结节良恶性是临床热点问题。传统超声依赖医师主观经验，诊断一致性欠佳。超声影像组学通过高通量提取医学图像中肉眼难以识别的定量特征，结合机器学习方法构建预测模型，为甲状腺结节良恶性鉴别提供了新的解决路径。本文系统综述超声影像组学的工作流程、不同超声模态(B-mode、弹性成像、超声造影等)的研究现状、多模态数据融合策略以及与深度学习结合的最新进展，评估其诊断效能与循证依据，分析当前面临的数据标准化、可重复性、临床转化等挑战，并对多中心前瞻性研究、可解释性AI及临床应用前景进行展望。总体而言，超声影像组学在甲状腺结节良恶性鉴别中展现出良好的诊断潜力，有望成为辅助临床决策的有效工具。

Abstract: The detection rate of thyroid nodules has been increasing annually, and non-invasive, accurate differentiation between benign and malignant nodules remains a clinical hotspot. Conventional ultrasound relies on subjective physician experience, leading to unsatisfactory diagnostic consistency. Ultrasound radiomics, by extracting high-throughput quantitative features imperceptible to the naked eye from medical images and combining them with machine learning models, offers a novel solution for distinguishing benign from malignant thyroid nodules. This review systematically summarizes the workflow of ultrasound radiomics, current research progress across different ultrasound modalities (B-mode, elastography, contrast-enhanced ultrasound, etc.), multimodal data fusion strategies, and recent advances combined with deep learning. We evaluate its diagnostic performance and evidence-based support, analyze challenges including data standardization, reproducibility, and clinical translation, and discuss prospects for multicenter prospective studies, explainable AI, and clinical application. Overall, ultrasound radiomics demonstrates promising diagnostic potential and may become an effective tool to aid clinical decision-making.