[Study on preoperative prediction of microvascular invasion based on contrast-enhanced ultrasound habitat imaging combined with peritumoral radiomics in hepatocellular carcinoma].

摘要

Objective: Hepatocellular carcinoma has a poor prognosis, and microvascular invasion (MVI) is an important indicator for guiding surgical decisions and influencing long-term survival. This study aims to explore the application value of habitat imaging combined with peritumoral radiomics in the preoperative assessment of MVI based on contrast-enhanced ultrasound (CEUS) images in hepatocellular carcinoma. Methods: The contrast-enhanced ultrasound images and clinical data of 186 patients with pathologically confirmed hepatocellular carcinoma at the First Affiliated Hospital of Guangxi Medical University were retrospectively analyzed and randomly divided into a training group (n=130) and a test group (n=56) in a 7∶3 ratio. Eleven clinical indicators, such as patient age and alpha-fetoprotein (AFP) were extracted. Microvascular invasion (MVI)-independent risk factors were screened to construct clinical prediction models for tumors. Peritumoral and intratumoral regions of interest (ROIs) were depicted at 5 mm and 10 mm on arterial and delayed CEUS phase images. Traditional radiomic features were extracted. Corresponding prediction models were constructed. The above-mentioned depicted intratumoral regions of interest were concurrently applied using K-means clustering to divide the tumor habitat subregions, habitat features, and establish an intratumoral habitat model. Furthermore, a combined model was constructed by integrating a habitat with 5 mm peritumoral features. The predictive efficacy of each model was comprehensively evaluated using indicators such as the area under the receiver operating characteristic (AUC), calibration curve, and decision curve. Results: The clinical model had poorly performed predictions for MVI (training set AUC=0.762; validation set AUC=0.668). The 5 mm peritumoral radiomics model showed better generalization ability and stability than the 10 mm peritumoral and intratumoral models (training set AUC=0.785 vs. 0.786, 0.802; validation set AUC=0.727 vs. 0.672, 0.708). The intratumoral habitat model (training set AUC=0.824; validation set AUC=0.753) was more effective in predicting MVI than both the clinical and the traditional radiomics models in hepatocellular carcinoma. Furthermore, the predictive efficacy of MVI with the selection of a constructed intratumoral habitat combined with 5 mm peritumoral was improved (training set AUC=0.888; test set AUC=0.843), with accuracy, sensitivity, specificity, and clinical net benefit all significantly higher than those of other models. The classification ability had a positive improvement effect. Conclusion: Intratumoral habitat analysis can effectively assess the MVI status based on contrast-enhanced ultrasound in patients with hepatocellular carcinoma. In addition, the combined 5-mm peritumoral radiomics model can more accurately assess the MVI status in patients with hepatocellular carcinoma. 目的： 基于超声造影图像，联合生境成像与瘤周影像组学，探索生境成像在肝细胞癌微血管浸润（MVI）术前评估中的应用价值。 方法： 回顾性分析2020年1月至2022年4月广西医科大学第一附属医院经病理证实为肝细胞癌的186例患者的超声造影图像资料及临床资料，按7∶3随机分为训练组（n=130）与测试组（n=56）。提取患者年龄、甲胎蛋白等11项临床指标，筛选MVI独立危险因素并构建临床预测模型；在超声造影动脉期、延迟期图像勾画瘤内、瘤周5 mm、瘤周10 mm感兴趣区，提取传统影像组学特征并构建对应预测模型；同时，利用上述已勾画的瘤内感兴趣区，采用K均值聚类划分肿瘤生境亚区域，提取生境特征并建立瘤内生境模型，进一步联合瘤周5 mm特征构建生境+瘤周5 mm联合模型。连续变量采用t检验或U检验，分类变量采用χ2检验或Fisher精确检验，以此分析训练组和测试组差异。通过单、多因素logistic回归筛选预测MVI的临床独立危险因素。通过受试者操作特征曲线下面积（AUC）、校准曲线、决策曲线等指标综合评估各模型预测效能。 结果： 临床模型预测MVI效果较差（训练组AUC=0.762；测试组AUC=0.668），瘤周5 mm影像组学模型的泛化能力与稳定性优于瘤内及瘤周10 mm模型（训练组AUC=0.785比0.786、0.802；测试组AUC=0.727比0.672、0.708），而瘤内生境模型（训练组AUC=0.824；测试组AUC=0.753）预测肝细胞癌的MVI效能优于临床模型与传统影像组学模型。因此，通过选择瘤内生境+瘤周5 mm构建联合模型，进一步提高了MVI的预测效能（训练组AUC=0.888，测试组AUC=0.843），准确度、敏感度、特异度及临床净获益均显著高于其他模型，且分类能力具有正向改善作用。 结论： 基于超声造影的瘤内生境分析可有效评估肝细胞癌MVI状态，联合瘤周5 mm影像组学模型能更准确地评估肝细胞癌患者MVI状态。.