肝细胞癌（hepatocellular carcinoma，HCC）具有空间异质性和时间异质性，既往研究表明肿瘤侵袭性前沿（tumor invasive front，TIF）作为肿瘤与宿主相互作用的区域可能最具恶性特征。肿瘤细胞的局部侵袭、转移与肿瘤不良预后最为相关。微转移同样也存在肿瘤与宿主相互作用的区域，仅仅考虑宿主与瘤体相接触的界面具有局限性。因此，我们对TIF概念进行了拓展，将肿瘤微转移纳入TIF。此外肝部分切除术后剩余肝脏作为微转移与肿瘤微环境的载体可能具有预后价值。本研究对TIF概念进行了拓展和完善，分为宏观TIF（macroscopic TIF，macTIF）和微观TIF（microscopic TIF，micTIF）。micTIF是包含了肿瘤微转移的区域。本研究在影像上将肿瘤边界内肉眼可见的5mm区域定义为macTIF，瘤周10mm区域定义为micTIF。本研究分别获取了瘤灶整体、肿瘤中心区、macTIF区域和micTIF区域高稳定性的增强CT影像组学特征并构建了预测无复发生存的预后模型，结果表明macTIF区域具有最好的预测效能并优于临床模型。macTIF区域影像组学特征的纳入提高了无复发生存的预测效能，并优于目前的分期系统。本研究发现瘤灶整体、肿瘤中心、macTIF和micTIF区域影像组学特征预测微转移的曲线下面积分别为0.780、0.784、0.823和0.752，结果显示macTIF 区域预测微转移效能最高，表明macTIF可能是微转移的主要贡献区域。本研究发现macTIF区域较肿瘤中心区E-钙黏蛋白表达下调、干细胞标志物EpCam表达上调，同时发现E-钙黏蛋白表达下调与微血管侵犯相关。这表明macTIF区域发生了上皮-间质转化，促进了侵袭转移。本研究发现影像组学特征能够预测E-钙黏蛋白的表达，曲线下面积为0.679。本研究首次发现余肝影像组学特征具有预后价值，并与肿瘤负荷和术后炎症状态相关。这表明术后残肝可能整合了炎症、肿瘤负荷等多种生物学信息。本研究首次用影像组学的方式阐明了macTIF区域是HCC中最有预后价值的区域，并发现macTIF区域发生了上皮-间质转化、富集了肿瘤干细胞。macTIF和micTIF能较好的预测微转移，这将有助于无瘤切缘的判断和辅助手术规划。CT影像组学特征能够较好的预测E-钙黏蛋白的表达，这提示我们可以用影像组学预测肿瘤生物标志物的表达，为治疗方案的选择、转换提供决策支持。本研究首次发现术后余肝影像组学特征具有预后价值，这可能能为随访中的临床决策提供动态、可量化的支持，提前预警肿瘤进展、精准干预。

Hepatocellular carcinoma (HCC) has spatial and temporal heterogeneity. Previous studies have shown that the tumor invasive front (TIF), as the region of tumor-host interaction, may be more malignant. The local invasion and metastasis of tumor cells are most associated with poor prognosis. Micrometastases also have interactions with the host. It is obviously not enough to only consider the interface between the host and the tumor. We incorporated micrometastases into TIF for the first time. Furthermore, the remanent liver after hepatectomy may have prognostic value as a carrier for micrometastases and the tumor microenvironment. In this study, we expanded and perfected the concept of TIF and divided it into macroscopic TIF (macTIF) and microscopic TIF (micTIF). The micTIF is the region that contains tumor micrometastases. From the imaging, the macTIF was defined as a 5 mm area within the tumor boundary, and the micTIF was defined as the 10 mm area around the tumor. The contrast-enhanced CT radiomic features of the whole tumor, tumor center, macTIF region and micTIF region with high stability were obtained, and a prognostic model for predicting disease-free survival was constructed. The results showed that the model based on the macTIF region had the best predictive performance, and better than clinical models. Inclusion of the macTIF regional radiomics features improved the predictive power of disease-free survival and was better than the current staging systems. In this study, it was found that the areas under the receiver operating characteristic curve of the radiomic features of the entire tumor, tumor center, macTIF and micTIF regions for predicting micrometastases were 0.780, 0.784, 0.823 and 0.752, respectively. This suggests that macTIF may be the main contributing region for micrometastases. This study found that the expression of E-cadherin was down-regulated and the expression of stem cell markers EpCam was up-regulated in the macTIF region compared with the central region of the tumor. It was also found that down-regulation of E-cadherin expression was associated with microvascular invasion.This indicated that epithelial-mesenchymal transition occurred in the macTIF region and promoted invasion and metastasis. This study showed that radiomic features could predict E-cadherin expression with an area under the curve of 0.679. For the first time, we found that radiomics features of remanent liver have prognostic value and correlate with tumor burden and postoperative inflammatory status. This suggests that the postoperative remanent liver may integrate various biological information such as inflammation and tumor burden.In this study, for the first time, radiomics was used to elucidate that macTIF was the most prognostic region in HCC, and it was found that the macTIF region underwent epithelial-mesenchymal transition and was enriched for cancer stem cells. macTIF and micTIF can better predict micrometastasis, which is helpful for the judgment of tumor-free resection margin and assisting surgical planning. CT radiomics features can predict the expression of E-cadherin, which suggests that we can use radiomics to predict the expression of tumor biomarkers and provide decision support for the selection and conversion of treatment options. For the first time, we found that radiomics features of postoperative remanent liver have prognostic value, which may provide dynamic and quantifiable support for clinical decision-making during follow-up, early warning of tumor progression, and precise intervention.