| 许昌华,何淼,王健.瘤内及瘤周CT影像组学联合临床特征预测早期肺腺癌淋巴血管浸润[J].中国医学影像技术,2024,40(10):1509~1513 |
| 瘤内及瘤周CT影像组学联合临床特征预测早期肺腺癌淋巴血管浸润 |
| Intratumoral and peritumoral CT radiomics combined with clinical features for predicting lymphovascular invasion of early lung adenocarcinoma |
| 投稿时间：2024-06-09 修订日期：2024-07-23 |
| DOI：10.13929/j.issn.1003-3289.2024.10.011 |
| 中文关键词: 肺肿瘤 腺癌 肿瘤转移 体层摄影术,X线计算机 影像组学 |
| 英文关键词:lung neoplasms adenocarcinoma neoplasm metastasis tomography,X-ray computed radiomics |
| 基金项目:重庆市技术创新与应用发展专项重点项目(CSTC2021jscx-gksb-N0025)。 |
|
| 摘要点击次数: 481 |
| 全文下载次数: 179 |
| 中文摘要: |
| 目的 观察基于瘤内及瘤周CT影像组学联合临床特征预测早期肺腺癌淋巴血管浸润(LVI)的价值。方法 回顾性分析252例Ⅰ~Ⅱa期肺腺癌,根据有无LVI将其分为阳性组(n=63)与阴性组(n=189),并按8∶2比例分为训练集(n=201)与测试集(n=51)。将组间差异有统计学意义的CT表现及临床资料纳入多因素logistic回归分析,筛选早期肺腺癌LVI的独立预测因素并构建CT-临床模型。利用自动方法及膨胀算法基于肿瘤及分别向外膨胀3、5、7 mm勾画ROI1~4,提取并筛选最优影像组学特征;分别以K近邻(KNN)、支持向量机(SVM)、决策树(DT)、随机森林(RF)及逻辑回归(LR)算法构建影像组学模型;筛选最佳分类器及可提供更多有效影像组学信息的ROI,以之构建最佳影像组学模型;基于CT-临床模型及最佳影像组学模型构建联合模型。绘制受试者工作特征曲线,计算曲线下面积(AUC),评估各模型预测早期肺腺癌LVI的效能。结果 实性病灶及CT-N+分期均为早期肺腺癌LVI的独立危险因素(P均＜0.05)。基于ROI3获得瘤内及瘤周5 mm总体积(GTPV5),以针对后者选出的14个最优特征构建的RFGTPV5模型为最佳影像组学模型。CT-临床模型、RFGPTV5模型及联合模型预测训练集早期肺腺癌LVI的AUC分别为0.875、0.908及0.917,在测试集分别为0.831、0.853及0.862。结论 瘤内及瘤周CT影像组学联合临床特征术前预测早期肺腺癌LVI的效能良好;基于瘤内+瘤周5 mm ROI可获得更多有价值的信息。 |
| 英文摘要: |
| Objective To observe the value of intratumoral and peritumoral CT radiomics combined with clinical features for preoperative predicting lymphovascular invasion (LVI) of early lung adenocarcinoma. Methods Totally 252 patients with stage Ⅰ—Ⅱa lung adenocarcinoma were retrospectively enrolled and were divided into positive group (n=63) and negative group (n=189) according to LVI or not, also into training set (n=201) and test set (n=51) at a ratio of 8∶2. Clinical data and CT findings being significantly different between groups were included in multivariate logistic regression analysis to screen independent predictors of early lung adenocarcinoma LVI and to construct CT-clinical model. The best radiomics features were extracted and screened in ROI of tumor (ROI1) and outward of 3(ROI2), 5 (ROI3) and 7 mm (ROI4) with automatic delineation and expanding algorithm, respectively. K-nearest neighbor (KNN), support vector machine (SVM), decision tree (DT), random forest (RF) and logistic regression (LR) algorithms were used to construct radiomics models. The best classifier algorithm and ROI which could provide more effective radiomics information were selected to construct the best radiomics model, which was used to construct the combined model combining with CT-clinical model. Receiver operating characteristic curves were drawn, and the areas under the curves (AUC) were calculated to evaluate the efficacy of the above models for preoperative predicting early lung adenocarcinoma LVI. Results Solid lesion and CT-N+ stage were both independent risk factors for early lung adenocarcinoma LVI (both P＜0.05). Peritumor 5 mm volume (GTPV5) was obtained based on ROI3, and the best radiomics model was the model established based on 14 optimal radiomics feature selected from RFGTPV5. AUC of CT-clinical model, RFGPTV5 model and combined model for preoperative predicting early lung adenocarcinoma LVI was 0.875, 0.908 and 0.917 in training set, while was 0.831, 0.853 and 0.862 in test set, respectively. Conclusion Intratumoral and peritumoral CT radiomics combined with clinical features had good efficacy for preoperative predicting LVI of early lung adenocarcinoma. Intratumor and peritumor 5 mm ROI could provide more valuable information. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
