基于计算机体层成像影像组学联合临床因素预测肝细胞癌患者经导管动脉化疗栓塞的近期疗效

作者： style="font-size: 12px ">邹彩霞 1 2 李静 2 黄刚 3 陈俊 4 赵建新 5 马娅琼 3 王莉莉 3

单位：1. 兰州市第二人民医院 核磁科 甘肃 兰州 730046 2. 甘肃中医药大学第一临床医学院 甘肃 兰州 730099 3. 甘肃省人民医院 放射科 甘肃 兰州 730099 4. 北京理工大学 北京 100081 5. 甘肃省肿瘤医院 影像科 甘肃 兰州 730000

关键词：肝细胞癌 经导管动脉化疗栓塞 影像组学 预测模型 治疗反应

分类号：

出版年,卷(期):页码：2026,18(1):50-61

摘要：

摘要：目的 构建并验证基于计算机体层成像（computed tomography，CT）影像组学 联合临床因素预测肝细胞癌（hepatocellular carcinoma，HCC）患者经导管动脉化疗栓 塞（transcatheter arterial chemoembolisation，TACE）治疗近期疗效的列线图模型。方法 回顾性分析兰州市第二人民医院 2020 年 8 月至 2023 年 11 月收治的 102 例接受 TACE 治疗的 HCC 患者的临床资料和 CT 图像，采用随机分层分组法将患者分为训练集 （81 例）和验证集（21 例）。TACE 治疗结束后，根据 RECIST1.1 评价标准确定靶肿瘤 的治疗反应。将完全缓解和部分缓解患者归为缓解组，将疾病稳定和疾病进展患者 归为未缓解组。使用 3Dslicer 软件在患者 TACE 术前 CT 平扫、动脉期、门静脉期及 TACE 术后 1 个月的 CT 平扫图像进行全肿瘤三维感兴趣区分割。使用 FAE 软件对图 像进行预处理，提取感兴趣区影像组学特征，计算术后平扫和术前平扫的影像组学特 征的变化值（Delta 影像组学特征）。通过最小绝对收缩和选择算子等方法从 Delta 影像 组学特征、动脉期、门静脉期影像组学特征中降维、筛选，分别构建预测 TACE 治疗 后 6 个月治疗反应的 Delta 影像组学模型、动脉期影像组学模型、门静脉期影像组学 模型并分别计算其组学评分。通过单因素以及后向多因素 Logistic 回归分析筛选临床 风险因子，并结合影像评分构建联合列线图。采用受试者工作特征（receiver operator characteristic，ROC）曲线评估模型的预测效能。采用校正曲线和决策曲线对模型的性 能进行评价。结果 训练集和验证集患者除有无瘤周动脉差异有统计学意义外（连续校 正 χ2 = 4.541，P = 0.033），其余指标差异均无统计学意义（P 均 ＞ 0.05）。训练集中 58 例缓解，23 例未缓解，验证集中 15 例缓解，6 例未缓解。训练集中缓解组和未缓解 组患者除天冬氨酸氨基转移酶（aspartate aminotransferase，AST）、白蛋白（albumin， Alb）、巴塞罗那肝癌临床分期（Barcelona clinic liver cancer staging，BCLC 分期）、远 处淋巴结转移比例差异有统计学意义外（P 均 ＜ 0.05），其余指标差异无统计学意 义（P 均 ＞ 0.05），验证集中缓解组和未缓解组患者除 AST、BCLC 分期、远处淋 巴结转移比例差异有统计学意义外（P 均 ＜ 0.05），其余指标差异均无统计学意义 （P 均 ＞ 0.05）。多因素 Logistic 回归分析表明，Alb（OR = 0.027，95%CI：0.012～0.418， P = 0.001）、远处淋巴结转移（OR = 0.771，95%CI：0.573～0.856，P = 0.001）、动脉期 影像组学特征评分（OR = 0.216，95%CI：0.096～0.336，P = 0.001）、静脉期影像组学 特征评分（OR = 0.197，95%CI：0.058～0.337，P = 0.005）及 Delta 特征值（OR = 0.226， 95%CI：0.116～0.336，P = 0.001）是 HCC 患者 TACE 治疗缓解的独立影响因素。使 用 Alb 和远处淋巴结转移构建临床模型，训练集和验证集 ROC 曲线下面积分别 0.865、0.778。利用动脉期影像组学特征评分、静脉期影像组学特征评分及 Delta 分别构建影像 组学模型，训练集上述指标的 ROC 曲线下面积分别为 0.750、0.692 和 0.709，验证集 分别为 0.711、0.622 和 0.789。联合 Alb、远处淋巴结转移、动脉期影像组学评分、门 静脉期影像组学评分、Delta 影像组学评分构建预测 HCC 患者 TACE 术后 6 个月缓解 的列线图模型，列线图在训练集和验证集的 ROC 曲线下面积分别为 0.965、0.889。列 线图模型的 ROC 曲线下面积均显著高于临床模型、动脉期影像组学评分、门静脉期影 像组学评分、Delta 影像组学评分（z = 2.85、P = 0.004，z = 2.62、P = 0.009，z = 2.47、 P = 0.013，z = 2.31、P = 0.021）。校准曲线显示所有校正曲线均与理想曲线有较好的 吻合度，提示列线图的准确性较高。临床决策曲线显示列线图的临床效益优于单独影 像组学评分。结论 基于 CT 影像组学联合临床资料构建的列线图能够预测 HCC 患者 TACE 术后 6 个月的治疗反应，可为临床制定精准个性化治疗方案提供新思路。

Abstract: Objective To construct and validate a nomogram model for predicting the short term therapeutic efficacy of transcatheter arterial chemoembolisation (TACE) in patients with hepatocellular carcinoma (HCC) based on computed tomography (CT) radiomics combined with clinical factors. Methods The clinical data and CT images of 102 patients with hepatocellular carcinoma (HCC) who received TACE treatment in Lanzhou Second People’s Hospital from August 2020 to November 2023 were retrospectively analyzed. The patients were divided into a training cohort (81 cases) and a validation cohort (21 cases) using the stratified random grouping method. After the completion of TACE treatment, the therapeutic response of target tumors was evaluated according to the RECIST version 1.1. Patients with complete response and partial response were classified into the response group, while those with stable disease and progressive disease were classified into the non-response group. Three-dimensional whole-tumor region of interest segmentation was performed on CT plain scan images, arterial phase, portal vein phase of patients before TACE and 1 month after TACE using the 3D Slicer software. Image preprocessing and radiomic feature extraction from the regions of interest were performed using the FAE software. Subsequently, the difference values of radiomic features of CT plain scan between post-TACE and pre-TACE, which were defined as Delta radiomic features, were calculated. Images were preprocessed and region-of-interest imaging histologic features were extracted using FAE software and change values of imaging histologic features (Delta imaging histologic features) were calculated for postoperative and preoperative scans. Dimensionality reduction and feature screening were performed on Delta radiomic features, arterial phase radiomic features and portal venous phase radiomic features using methods including the least absolute shrinkage and selection operator. Three radiomic models for predicting the 6-month therapeutic response after  TACE, namely the Delta radiomics model, arterial phase radiomics model, and portal venous phase radiomics model, were constructed respectively, and the corresponding radiomics scores of each model were calculated separately. Clinical risk factors were screened through univariate analysis and subsequent backward stepwise multivariate Logistic regression, and a combined nomogram was constructed by integrating the radiomics scores. The predictive efficacy of the models was evaluated using receiver operator characteristic (ROC) curves. The performance of the model was further assessed via calibration curves and decision curve analysis. Results No statistically significant differences were observed in all indicators between patients in the training cohort and the validation cohort (all P ＞ 0.05), except for the presence or absence of peritumoral artery (continuity-corrected χ2 = 4.541, P = 0.033). In the training cohort, 58 patients were classified as responders and 23 as non-responders; in the validation cohort, 15 patients were classified as responders and 6 as non-responders. For the training cohort, there were statistically significant differences in aspartate aminotransferase (AST), albumin (Alb), Barcelona clinic liver cancer staging (BCLC stage) and proportion of distant lymph node metastasis of patients in response group and non-response group (all P ＜ 0.05), while no statistically significant differences were found in other indicators (all P ＞ 0.05). For the validation cohort, statistically significant differences between the two groups were observed in AST, BCLC stage and proportion of distant lymph node metastasis (all P ＜ 0.05), and no statistically significant differences were detected in other indicators (all P ＞ 0.05). Multivariate Logistic regression analysis showed that Alb (OR = 0.027, 95%CI: 0.012～0.418, P = 0.001), distant lymph node metastasis (OR = 0.771, 95%CI: 0.573～0.856, P = 0.001), arterial phase radiomic score (OR = 0.216, 95%CI: 0.096～0.336, P = 0.001), portal venous phase radiomic score (OR = 0.197, 95%CI: 0.058～0.337, P = 0.005) and Delta radiomic feature value (OR = 0.226, 95%CI: 0.116～0.336, P = 0.001) were independent affecting factors for therapeutic response to TACE in patients with HCC. A clinical model was constructed using Alb and distant lymph node metastasis, with areas under the ROC curve of 0.865 and 0.778 in the training cohort and the validation cohort, respectively. Three separate radiomic models were developed based on the arterial phase radiomic score, portal venous phase radiomic score, and Delta radiomic feature value, respectively. The areas under the ROC curve of the above three models were 0.750, 0.692 and 0.709 in the training cohort, and 0.711, 0.622 and 0.789 in the validation cohort, respectively. A nomogram model for predicting 6-month therapeutic response after TACE in HCC patients was constructed by integrating Alb, distant lymph node metastasis, arterial phase radiomic score, portal venous phase radiomic score, and Delta radiomic score. The areas under the ROC curve of the nomogram were 0.965 and 0.889 in the training cohort and validation cohort, respectively. The area under the ROC curve of the nomogram model was significantly higher than those of the clinical model, arterial phase radiomic model, portal venous phase radiomic model and Delta radiomic model, respectively (z = 2.85, P = 0.004; z = 2.62, P = 0.009; z = 2.47, P = 0.013; z = 2.31, P = 0.021). Calibration curves showed good agreement between all fitted calibration lines and the ideal perfect prediction line, suggesting high predictive accuracy of the nomogram. Decision curve analysis results indicated that the nomogram achieved a higher net clinical benefit than individual radiomic scores. Conclusion The model constructed based on CT imaging histology combined with clinical data was able to predict the 6-month treatment response to TACE in patients with HCC, and could provide a new idea for the clinical development of precise and personalized treatment plans.

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】