| 陈巧玲,冯峰,李曼曼,薛婷,彭慧,石健,朱兴华,段绍峰.联合含瘤周肿瘤全体积CT影像组学特征及临床相关独立预测因子列线图预测肺腺癌淋巴血管侵犯[J].中国医学影像技术,2022,38(8):1181~1186 |
| 联合含瘤周肿瘤全体积CT影像组学特征及临床相关独立预测因子列线图预测肺腺癌淋巴血管侵犯 |
| Nomogram based on CT radiomics features of gross peritumoral tumor volume and clinical relevant independent predictors for predicting lymphovascular invasion of lung adenocarcinoma |
| 投稿时间：2022-02-15 修订日期：2022-05-05 |
| DOI：10.13929/j.issn.1003-3289.2022.08.013 |
| 中文关键词: 肺肿瘤 体层摄影术,X线计算机 影像组学 淋巴血管侵犯 |
| 英文关键词:lung neoplasms tomography, X-ray computed radiomics lymphovascular invasion |
| 基金项目:南通市2021年度市级社会民生科技项目(MS22021047)。 |
|
| 摘要点击次数: 1577 |
| 全文下载次数: 347 |
| 中文摘要: |
| 目的 观察基于含瘤周的肿瘤全体积(GPTV) CT影像组学特征及临床相关独立预测因子构建的联合模型列线图预测肺腺癌淋巴血管侵犯(LVI)的价值。方法 回顾性分析142例经病理证实的肺腺癌患者,以7 ∶ 3比例将其随机分为训练集(n=100,40例LVI阳性、60例LVI阴性)和验证集(n=42,17例LVI阳性、25例LVI阴性)。以单因素分析及多因素logistic回归分析筛选肺腺癌LVI的临床相关独立预测因子,以之构建临床模型。分别基于肿瘤全体积(GTV)及含瘤周3 mm、6 mm、9 mm的GPTV (GPTV3、GPTV6和GPTV9)的增强动脉期CT图提取并筛选最佳影像组学特征,构建影像组学模型,即GTV、GPTV3、GPTV6和GPTV9模型并筛选最佳者;基于后者的影像组学评分和临床相关独立预测因子构建联合模型,绘制列线图进行可视化。以受试者工作特征(ROC)曲线评估各模型预测肺腺癌LVI的效能,以决策曲线分析(DCA)评价联合模型列线图的价值。结果 性别、吸烟和毛刺征均为肺腺癌LVI的临床相关独立预测因子(P均＜0.05)。分别基于GTV、GPTV3、GPTV6及GPTV9筛选出7、16、10及8个最佳影像组学特征,用于构建GTV、GPTV3、GPTV6及GPTV9模型。GPTV3模型预测训练集、验证集肺腺癌LVI的曲线下面积(AUC)分别为0.82、0.77,均高于GTV (0.79、0.72,Z=3.74、2.62,P均＜0.01)、GPTV6(0.80、0.72,Z=2.40、2.06,P均＜0.05)及GPTV9(0.77,0.72,Z=3.03、2.59,P均＜0.01),为最佳影像组学模型。联合模型列线图(0.86、0.73,Z=2.66、2.31,P均＜0.05)及GPTV3模型(0.82、0.77,Z=2.23、2.54,P均＜0.05)于训练集和验证集的AUC均高于临床模型(0.73、0.61),而联合模型列线图与GPTV3模型的AUC差异均无统计学意义(Z=1.57、0.88,P均＞0.05)。阈值取0.20~0.50时,联合模型列线图与GPTV3模型的净获益相当,且均大于临床模型。结论 基于GPTV3影像组学特征及临床相关独立预测因子的列线图可有效预测肺腺癌LVI。 |
| 英文摘要: |
| Objective To explore the value of the nomogram based on CT radiomics features gross peritumoral tumor volume (GPTV) and clinical relevant independent predictors for predicting lymphovascular invasion (LVI) of lung adenocarcinoma. Methods Data of 142 patients with pathologically confirmed lung adenocarcinoma were retrospectively analyzed. The patients were randomly divided into training set (n=100, including 40 LVI-positive and 60 LVI-negative ones) and validation set (n=42, including 17 LVI-positive and 25 LVI-negative) at the ratio of 7:3. Univariate analysis and multivariate logistic regression analysis were used to select clinical relevant independent predictors for LVI of lung adenocarcinoma to construct the clinical model. Based on enhanced CT arterial phase images of gross tumor volume (GTV) and GPTV incorporating peritumoral 3 mm, 6 mm and 9 mm regions (GPTV3, GPTV6 and GPTV9), the best radiomics features were extracted and screened to construct radiomics models, including GTV, GPTV3, GPTV6 and GPTV9 models. Then the best radiomics model was selected. A combined model was constructed based on radiomics score of the best radiomics model and clinical relevant independent predictors, and a nomogram was drawn for visualization of the combined model. Receiver operating characteristic (ROC) curves were drawn to evaluate the efficacy of each model for predicting LVI of lung adenocarcinoma, and decision curve analysis (DCA) was used to assess the value of the combined model nomogram. Results Gender, smoking and spiculation were all clinically relevant independent predictors for LVI of lung adenocarcinoma (all P<0.05), and were used to establish the clinical model. Based on GTV, GPTV3, GPTV6 and GPTV9, 7, 16, 10 and 8 optimal radiomics features were selected, respectively, and GTV, GPTV3, GPTV6 and GPTV9 models were constructed. The areas under the curve (AUC) of GPTV3 model for predicting LVI of lung adenocarcinoma in the training set and validation set was 0.82 and 0.77, respectively, all higher than those of GTV (0.79, 0.72, Z=3.74, 2.62, both P<0.01), GPTV6 (0.80, 0.72, Z=2.40, 2.06, both P<0.05) and GPTV9 (0.77, 0.72, Z=3.03, 2.59, both P<0.01), and GPTV3 model was selected as the best radiomics model. AUC of the combined model nomogram (0.86, 0.73, Z=2.66, 2.31, both P<0.05) and GPTV3 model (0.82, 0.77, Z=2.23,2.54, both P<0.05) in the training set and validation set were higher than those of the clinical model (0.73, 0.61), while there was no significant difference of AUC of the combined model nomogram nor GPTV3 model (Z=1.57, 0.88, both P>0.05). Taken 0.20-0.50 as the threshold of probability, the net benefits of the combined model nomogram and GPTV3 model were similar, both larger than that of the clinical model. Conclusion Nomogram based on GPTV3 radiomics features and clinical relevant independent predictors could effectively predict LVI of lung adenocarcinoma. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
