基于超声影像组学模型对肝细胞癌肿瘤分化等级的评估价值

作者：郭明珍  王俪洁  唐佳盈

单位：广元市第一人民医院 超声医学科 四川 广元 628000

关键词：：肝细胞癌 肿瘤分化等级 超声影像组学 LASSO回归算法 列线图预测模型

分类号：

出版年,卷(期):页码：2024,16(3):8-16

摘要：

摘要：目的 探究肝细胞癌（hepatocellular carcinoma，HCC）不同分化等级患者的超声 影像组学特征及其对肿瘤分化等级的诊断价值。方法 选择2021年3月至2023年3月于广 元市第一人民医院就诊并经穿刺活检或手术证实为HCC的212例患者为研究对象纳入训 练集，根据肿瘤分化等级将患者分为高分化组（Ⅰ、Ⅱ级，138例）和低分化组（Ⅲ、 Ⅳ级，74例），比较两组患者的临床资料，包括性别、年龄、吸烟、饮酒、肝癌家族 史、肿瘤直径、病变部位、临床分期、Child-Pugh分级、乙型肝炎、肝硬化、肿瘤包 膜、淋巴结肿大、天门冬氨酸氨基转移酶（aspartate aminotransferase，AST）、丙氨酸 氨基转移酶（alanine aminotransferase，ALT）、总胆红素、白蛋白、血小板、Ki-67、 甲胎蛋白（alpha fetoprotein，AFP）。按照相同纳入与排除标准另选取本院同期收治 的60例HCC患者纳入验证集，用于模型的外部验证。采用多因素Logistic回归分析患者 肿瘤低分化的影响因素；采集患者的超声图像并提取影像组学特征，采用LASSO回归 算法筛选与肿瘤分化等级高度相关的超声影像组学特征（F），并获得其系数（α）； 采用受试者工作特征（receiver operator characteristic，ROC）曲线分析临床参数模型、 超声影像组学评分模型及联合模型的效能；采用R软件构建预测HCC患者肿瘤低分化 的列线图模型，采用ROC曲线评价列线图模型的区分度，分别采用校准曲线和临床决 策曲线评价列线图模型的准确性和有效性。结果 Logistic回归分析表明肝硬化（OR = 1.720，95%CI：1.183～2.311，P = 0.010）、血小板≥ 183.69 × 109/L（OR = 1.418， 95%CI：1.051～1.932，P = 0.025）、Ki-67阳性（OR = 1.552，95%CI：1.363～1.770， P = 0.017）、AFP阳性（OR = 2.021，95%CI：1.230～2.786，P ＜ 0.001）是HCC 患者肿瘤低分化的危险因素，AST ≥ 55.14 U/L为保护因素（OR = 0.511，95%CI： 0.119～0.878，P = 0.002）。经LASSO回归算法共筛选出9个超声影像组学特征，超 声影像组学评分=－1.071 + 。训练集中临床参数模型、超声影像组学评分模 型及联合模型的ROC曲线下面积分别为0.702（95%CI：0.638～0.775，P ＜ 0.001）、 0.805（95%CI：0.814～0.893，P ＜ 0.001）和0.914（95%CI：0.846～0.972，P ＜ 0.001）；验证集中3种模型的ROC曲线下面积分别为0.712（95%CI：0.659～0.782， P ＜ 0.001）、0.793（95%CI：0.745～0.839，P ＜ 0.001）和0.895（95%CI： 0.846～0.951，P ＜ 0.001）；训练集和验证集超声影像组学评分模型均优于临床参数 模型（z = 2.502、2.475，P = 0.024、0.031），且联合模型的预测性能更优（z = 2.782、 2.686，P = 0.011、0.018）。训练集和验证集列线图模型的ROC曲线下面积分别为0.914 （95%CI：0.873～0.955，P ＜ 0.001）和0.905（95%CI：0.836～0.934，P ＜ 0.001）， 敏感度分别为89.72%和85.43%，特异度分别为87.24%和80.58%，列线图模型的区分度 较好；训练集和验证集列线图模型预测HCC患者肿瘤低分化风险与实际观测值拟合良 好（Hosmer-Lemeshow检验χ2 = 2.502、2.388，P = 0.112、0.096），模型预测的准确性 较好；训练集和验证集中，高风险阈值在0.01～1.00内时列线图模型曲线的净获益值 均＞ 0，且远离肿瘤低分化参考线和肿瘤高分化参考线2条极端曲线，表明列线图模型 对HCC患者肿瘤分化等级鉴别的临床应用价值较高，有效性较好，实用性较强。结论 超声影像组学特征与HCC肿瘤分化等级密切相关，超声影像组学模型可用于预测HCC 患者肿瘤低分化状态。

Abstract: Objective To explore the ultrasonographic features of hepatocellular carcinoma (HCC) patients with different differentiation grades and its diagnostic value on poor differentiation of tumors. Methods A total of 212 patients with HCC confirmed by biopsy or surgery in the First People’s Hospital of Guangyuan from March 2021 to March 2023 were selected as the research objects and included in the training set. According to the tumor differentiation grade, the patients were divided into well-differentiated group (grade Ⅰ and Ⅱ, 138 cases) and poorly differentiated group (grade Ⅲ and Ⅳ, 74 cases). The clinical data of patients in two groups were compared, including gender, age, smoking, drinking, family history of HCC, tumor diameter, lesion location, clinical stage, Child-Pugh classification, hepatitis B, liver cirrhosis, tumor capsule, lymph node enlargement, aspartate aminotransferase (AST), alanine aminotransferase (ALT), total bilirubin, albumin, platelet, Ki-67 and alpha fetoprotein (AFP). According to the same inclusion and exclusion criteria, 60 HCC patients admitted to our hospital during the same period were selected into the validation set for external validation of the model. Multivariate Logistic regression was used to analyze the influencing factors of poor tumor differentiation. The ultrasound images of the patients were collected and the radiomics features were extracted. The LASSO regression algorithm was used to screen the ultrasound radiomics features (F) which were highly related to the tumor differentiation grade, and its coefficient (α) was obtained. The receiver operator characteristic (ROC) curve was used to analyze the efficacy of the clinical parameter model, the ultrasound radiomics score model and the combined model. R software was used to construct a nomogram model for predicting poor differentiation of HCC patients. ROC curve was used to evaluate the discrimination of the nomogram model, and calibration curve and clinical decision curve were used to evaluate the accuracy and effectiveness of the nomogram model. Results Logistic regression analysis showed that liver cirrhosis (OR = 1.720, 95%CI: 1.183～2.311, P = 0.010), platelets ≥ 183.69 × 109/L (OR = 1.418, 95% CI: 1.051～1.932, P = 0.025), Ki-67 positive (OR = 1.552, 95%CI: 1.363～1.770, P = 0.017) and AFP positive (OR = 2.021, 95%CI: 1.230～2.786, P ＜ 0.001) were risk factors for poorly differentiated tumors, and AST ≥ 55.14 U/L (OR = 0.511, 95%CI: 0.119～0.878, P = 0.002) was a protective factor. A total of 9 ultrasound imaging characteristics were screened out through the LASSO regression algorithm, and the ultrasound imaging score =－1.071 + . The areas under the ROC curve for the clinical parameter model, the ultrasound radiomics score model and the combined model in training set were 0.702 (95%CI: 0.638～0.775, P ＜ 0.001), 0.805 (95%CI: 0.814～0.893, P ＜ 0.001) and 0.914 (95%CI: 0.846～0.972, P ＜ 0.001), respectively. The areas under the ROC curve for the three models in the validation set were 0.712 (95%CI: 0.659～0.782, P ＜ 0.001), 0.793 (95%CI: 0.745～0.839, P ＜ 0.001) and 0.895 (95%CI: 0.846～0.951, P ＜ 0.001), respectively. The ultrasound imaging histological scoring models in the training set and verification set were better than the clinical parameter models (z = 2.502, 2.475, P = 0.024, 0.031), and the combined model had better predictive performance (z = 2.782, 2.686, P = 0.011, 0.018). The areas under the ROC curve of the nomogram model for the training set and validation set were 0.914 (95%CI: 0.873～0.955, P ＜ 0.001) and 0.905 (95%CI: 0.836～0.934, P ＜ 0.001), respectively, the sensitivities were 89.72% and 85.43%, respectively, and the specificities were 87.24% and 80.58%, respectively. The nomogram model had good discrimination. The nomogram models of the training set and validation set on predicting the risk of tumor poor differentiation in HCC patients fit well with the actual observed values (Hosmer-Lemeshow test χ2 = 2.502, 2.388, P = 0.112, 0.096), and the accuracy was good. In both the training set and the verification set, when the high-risk threshold was within the range of 0.01 to 1.00, the net benefit values of the nomogram model curves were all ＞ 0, and were far away from the two extreme curves of the low differentiation reference line and the high differentiation reference line of the tumor. This showed that the nomogram model had high clinical application value, good effectiveness and strong practicality in identifying tumor differentiation levels in patients with HCC. Conclusions The characteristics of ultrasound imaging were closely related to the grade of tumor differentiation in HCC. Ultrasound imaging model can be used to predict the state of poor differentiation in patients with HCC.

基金项目：

作者简介：

参考文献：

服务与反馈：

【文章下载】【加入收藏】