影像组学在甲状腺恶性肿瘤中的研究进展

Research Progress of Radiomics in Thyroid Malignant Tumors

摘要: 目的：综述影像组学在甲状腺恶性肿瘤的诊断分层、侵袭性评估、转移预测与疗效监测中的研究进展。方法：系统检索国内外数据库，纳入影像组学/深度学习研究、系统综述与方法学指南，并对影像获取、分割、特征工程、验证及临床效益评估进行综合分析。结果：证据显示，超声与CT影像组学在结节良恶性鉴别、中央区隐匿性淋巴结转移预测、腺体外侵犯与分子风险提示等任务中具有一定增益；多中心外部验证研究提示平扫CT与增强特征可提供可迁移信息，但不同中心协议差异仍会导致性能衰减。同时，透明报告与偏倚控制框架有助于提高研究可信度。结论：影像组学有望成为细针穿刺与常规影像的有效补充，但需在标准化流程、多中心前瞻性验证、校准与决策收益评估方面进一步完善。

Abstract: Objective: To review recent advances in radiomics for diagnostic stratification, assessment of tumor aggressiveness, prediction of metastasis, and treatment response monitoring in malignant thyroid tumors. Methods: A systematic search of domestic and international databases was conducted to identify radiomics/deep learning studies, systematic reviews, and methodological guidelines. A comprehensive synthesis was performed focusing on imaging acquisition, segmentation, feature engineering, validation strategies, and evaluation of clinical utility. Results: The available evidence indicates that ultrasound- and CT-based radiomics provides incremental value in tasks such as differentiating benign from malignant nodules, predicting occult central compartment lymph node metastasis, assessing extrathyroidal extension, and inferring molecular risk. Multicenter external validation studies suggest that non-contrast CT and contrast-enhanced features can yield transferable information; however, inter-center protocol heterogeneity remains associated with performance degradation. Transparent reporting practices and bias-control frameworks were shown to enhance study credibility. Conclusions: Radiomics has the potential to serve as an effective adjunct to fine-needle aspiration and conventional imaging. Nevertheless, further refinement is required in terms of standardized workflows, multicenter prospective validation, model calibration, and assessment of decision-analytic benefit.