| 林苗苗,李凯,黄海鹏,赵祥.瘤内及瘤周影像组学联合临床和CT特征预测浸润性肺腺癌周围脉管侵犯[J].中国医学影像技术,2023,39(10):1514~1519 |
| 瘤内及瘤周影像组学联合临床和CT特征预测浸润性肺腺癌周围脉管侵犯 |
| Intratumoral and peritumoral radiomics combined with clinical and CT features for predicting peritumoral lymphovascular invasion of pulmonary invasive adenocarcinoma |
| 投稿时间：2023-05-28 修订日期：2023-08-18 |
| DOI：10.13929/j.issn.1003-3289.2023.10.015 |
| 中文关键词: 肺肿瘤 体层摄影术,X线计算机 肿瘤转移 影像组学 |
| 英文关键词:lung neoplasms tomography, X-ray computed neoplasm metastasis radiomics |
| 基金项目:广西医疗卫生适宜技术开发与推广应用项目(S2020036)。 |
|
| 摘要点击次数: 2048 |
| 全文下载次数: 707 |
| 中文摘要: |
| 目的 观察瘤内及瘤周影像组学联合临床和CT特征预测浸润性腺癌(IAC)周围脉管侵犯(LVI)的价值。方法 回顾性分析190例经术后病理确诊IAC患者,其中65例LVI(+)、125例LVI(－),按照7∶3比例将其分为训练集 [134例,46例瘤周LVI(+)、88例瘤周LVI(－)]和测试集 [56例,19例瘤周LVI(+)、37例瘤周LVI(－)]。对临床及CT表现行单因素及多因素logistic分析,筛选IAC瘤周LVI(+)的独立预测因素,构建临床-CT模型。分别基于瘤灶(GT)、肿瘤-瘤周过渡区(GPT)及瘤周区(PT)提取影像组学特征,以其最佳者构建影像组学模型(模型GT、模型GPT和模型PT),并筛选最佳影像组学模型;基于最佳模型影像组学评分与临床、CT独立预测因子构建联合模型,并绘制列线图。绘制受试者工作特征(ROC)曲线,计算曲线下面积(AUC),评估各模型预测IAC瘤周LVI的效能;以校准曲线评估模型校准度,以决策曲线分析评价模型的临床价值。结果 年龄、术前癌胚抗原水平及毛刺征均为IAC瘤周LVI(+)的独立预测因子(P均＜0.05)。模型GPT为最佳影像组学模型。联合模型在训练集的AUC高于临床-CT模型(0.90 vs.0.82,P=0.01)及模型GTP(0.90 vs. 0.83,P＜0.01),在测试集中的AUC高于临床-CT模型(0.87 vs.0.74,P=0.01)而与模型GTP差异无统计学意义(0.87 vs.0.79,P=0.20)。在训练集及测试集中,临床-CT模型与模型GTP的AUC差异均无统计学意义(P均＞0.05)。临床-CT模型、模型GPT及联合模型的校准度均较高。阈值取0.01~0.90时,联合模型的临床净效益更大。结论 瘤内及瘤周影像组学联合临床和CT特征可有效预测IAC瘤周LVI。 |
| 英文摘要: |
| Objective To observe the value of intratumoral and peritumoral lymphovascular combined with clinical and CT features for predicting peritumoral lymphovascular invasion (LVI) of pulmonary invasive adenocarcinoma (IAC). Methods Data of 190 patients with IAC confirmed by postoperative pathology, including 65 cases with LVI (+) and 125 cases with LVI (－) were retrospectively analyzed. The patients were divided into training set (n=134, including 46 LVI [+] and 88 LVI [-]) and test set (n=56, including 19 LVI [+] and 37 LVI [-]) at the ratio of 7∶3. Univariate and multivariate logistic analysis were used to analyze clinical and CT findings, so as to screen the independent predictors of IAC peritumoral LVI (+) to construct the clinical-CT model. The best radiomics features of gross tumor (GT), gross tumor and peritumor (GPT) and peritumor (PT) areas were extracted and screened to construct radiomics models (modelGT, modelGPT and modelPT), and the optimal radiomics model was screened. A combined model was constructed based on the optimal model radiomics score, clinical and CT independent predictors, and its nomogram was drawn. Receiver operating characteristic (ROC) curves were drawn, and the area under the curve (AUC) was calculated to evaluate the efficacy of each model for predicting peritumoral LVI of IAC. The calibration curve was used to evaluate the calibration degree of each model, and the clinical value of the models were evaluated using decision curve analysis (DCA). Results Age, preoperative CEA level and spiculation were independent predictors of peritumoral LVI (+) of IAC (all P＜0.05). ModelGPT was considered as the optimal radiomics model. In training set, the AUC of combined model was higher than that of clinical-CT model (0.90 vs. 0.82, P=0.01) and modelGTP (0.90 vs. 0.83, P＜0.01), while in test set the AUC of combined model was higher than that of clinical-CT model (0.87 vs. 0.74, P＜0.01) but not significant different with AUC of modelGTP (0.87 vs. 0.79, P=0.20). In both training set and test set, no significant difference of AUC was found between clinic-CT model and modelGPT (both P＞0.05). The calibration degree of clinical-CT model, modelGPT and combined model were all high. The clinical net benefit of combined model was great at the threshold of 0.01－0.90. Conclusion Intratumoral and peritumoral radiomics combined with clinical and CT features could be used to effectively predict peritumoral LVC of IAC. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
