Zeng Chao, Zhang Wei, Liu Meiyue, Liu Jianping, Zheng Qiangxin, Li Jianing, Wang Zhiwu, Sun Guogui
Hebei Key Laboratory of Medical-industrial Integration Precision Medicine, Clinical Medicine College, Affiliated Hospital, North China University of Science and Technology, Tangshan, Hebei, China.
Department of Radiotherapy, Yantai Yuhuangding Hospital, The Affiliated Hospital of Qingdao University, Yantai, Shandong, China.
Front Oncol. 2023 May 24;13:1096364. doi: 10.3389/fonc.2023.1096364. eCollection 2023.
This study aimed to establish a predictive model for occult lymph node metastasis (LNM) in patients with clinical stage I-A non-small cell lung cancer (NSCLC) based on contrast-enhanced CT.
A total of 598 patients with stage I-IIA NSCLC from different hospitals were randomized into the training and validation group. The "Radiomics" tool kit of AccuContour software was employed to extract the radiomics features of GTV and CTV from chest-enhanced CT arterial phase pictures. Then, the least absolute shrinkage and selection operator (LASSO) regression analysis was applied to reduce the number of variables and develop GTV, CTV, and GTV+CTV models for predicting occult lymph node metastasis (LNM).
Eight optimal radiomics features related to occult LNM were finally identified. The receiver operating characteristic (ROC) curves of the three models showed good predictive effects. The area under the curve (AUC) value of GTV, CTV, and GTV+CTV model in the training group was 0.845, 0.843, and 0.869, respectively. Similarly, the corresponding AUC values in the validation group were 0.821, 0.812, and 0.906. The combined GTV+CTV model exhibited a better predictive performance in the training and validation group by the Delong test (<0.05). Moreover, the decision curve showed that the combined GTV+CTV predictive model was superior to the GTV or CTV model.
The radiomics prediction models based on GTV and CTV can predict occult LNM in patients with clinical stage I-IIA NSCLC preoperatively, and the combined GTV+CTV model is the optimal strategy for clinical application.
本研究旨在基于增强CT建立临床I - A期非小细胞肺癌（NSCLC）患者隐匿性淋巴结转移（LNM）的预测模型。
将来自不同医院的598例I - IIA期NSCLC患者随机分为训练组和验证组。采用AccuContour软件的“放射组学”工具包从胸部增强CT动脉期图像中提取GTV和CTV的放射组学特征。然后，应用最小绝对收缩和选择算子（LASSO）回归分析减少变量数量，并建立用于预测隐匿性淋巴结转移（LNM）的GTV、CTV和GTV + CTV模型。
最终确定了8个与隐匿性LNM相关的最佳放射组学特征。三个模型的受试者工作特征（ROC）曲线显示出良好的预测效果。训练组中GTV、CTV和GTV + CTV模型的曲线下面积（AUC）值分别为0.845、0.843和0.869。同样，验证组中的相应AUC值分别为0.821、0.812和0.906。通过德龙检验（<0.05），联合GTV + CTV模型在训练组和验证组中表现出更好的预测性能。此外，决策曲线表明联合GTV + CTV预测模型优于GTV或CTV模型。
基于GTV和CTV的放射组学预测模型可在术前预测临床I - IIA期NSCLC患者的隐匿性LNM，联合GTV + CTV模型是临床应用的最佳策略。
