肝癌瘤内与瘤周超声影像组学联合分析的 临床应用价值与挑战

Clinical Application Value and Challenges of Combined Analysis of Intratumoral and Peritumoral Ultrasound Radiomics in Hepatocellular Carcinoma

摘要: 肝细胞癌(HCC)发病率持续上升且预后不良，早期诊断与精准治疗需求迫切，传统影像学在评估肿瘤异质性、预测治疗反应等方面存在显著局限。影像组学作为定量图像分析技术，为肝癌无创精准诊疗提供了新路径，其中瘤内与瘤周超声影像组学联合分析因能捕捉肿瘤空间异质性及瘤周微环境生物学信息，成为研究热点。本文系统梳理肝癌影像组学在特征提取、多模态数据融合、瘤内与瘤周区域界定等方面的技术进展，重点分析联合分析在微血管侵犯预测、免疫评分优化、早期复发风险评估及局部联合系统治疗反应预测中的临床价值，总结瘤周3~5 mm特征提取、机器学习算法优化、多中心数据验证及临床–影像组学联合建模等关键技术方法。研究表明，联合模型较单一区域模型可显著提升MVI预测(AUC 0.80~0.85 vs. 0.70~0.75)、免疫治疗响应预测(AUC提升约20%)、早期复发评估(C-index 0.82，较TNM分期提升0.15)等效能，瘤周特征对治疗耐药预测贡献率达40%。但当前研究仍面临瘤周区域界定缺乏标准化、不同影像模态特征可比性差、模型可解释性不足、多中心数据异质性难处理等核心挑战。未来需通过建立统一界定标准、开发可解释性算法、优化多中心数据整合策略等推动技术临床转化，助力肝癌精准医疗体系的构建与完善。

Abstract: Hepatocellular carcinoma (HCC) is characterized by a rising incidence and poor prognosis, creating an urgent clinical need for early diagnosis and precision treatment. Traditional imaging modalities have significant limitations in assessing tumor heterogeneity and predicting therapeutic response. As a quantitative image analysis technique, radiomics provides a new approach for non-invasive and precise diagnosis and treatment of HCC. Among these, the combined analysis of intratumoral and peritumoral ultrasound radiomics has emerged as a research hotspot due to its ability to capture tumor spatial heterogeneity and biological information of the peritumoral microenvironment. This paper systematically reviews the technical advances of HCC radiomics in feature extraction, multimodal data fusion, and the delineation of intratumoral and peritumoral regions. It focuses on analyzing the clinical value of combined analysis in predicting microvascular invasion (MVI), optimizing immune scoring, assessing early recurrence risk, and predicting response to local combined systemic therapy, and summarizes key technical methods such as feature extraction from the 3~5 mm peritumoral region, optimization of machine learning algorithms, multicenter data validation, and construction of clinical-radiomics combined models. Studies have shown that combined models can significantly improve the predictive efficacy of MVI (AUC 0.80~0.85 vs. 0.70~0.75), immunotherapy response (AUC increased by approximately 20%), and early recurrence assessment (C-index 0.82, 0.15 higher than TNM staging) compared with single-region models. Peritumoral features contribute up to 40% to the prediction of therapeutic resistance. However, current research still faces core challenges including the lack of standardized delineation of peritumoral regions, poor comparability of features across different imaging modalities, insufficient model interpretability, and difficulties in handling multicenter data heterogeneity. In the future, it is necessary to promote the clinical translation of this technology by establishing unified delineation standards, developing interpretable algorithms, and optimizing multicenter data integration strategies, thereby facilitating the construction and improvement of the precision medicine system for HCC.