Huang Xiaoni, Xue Yang, Deng Bing, Chen Jun, Zou Jiani, Tan Huibin, Jiang Yuanliang, Huang Wencai
The First School of Clinical Medicine, Southern Medical University, Guangzhou, China.
Department of Radiology, General Hospital of Central Theater Command of the People's Liberation Army, Wuhan, China.
Front Oncol. 2024 Sep 27;14:1406166. doi: 10.3389/fonc.2024.1406166. eCollection 2024.
To investigate the value of CT radiomics combined with radiological features in predicting pathological grade of stage I invasive pulmonary adenocarcinoma (IPA) based on the International Association for the Study of Lung Cancer (IASLC) new grading system.
The preoperative CT images and clinical information of 294 patients with stage I IPA were retrospectively analyzed (159 training set; 69 validation set; 66 test set). Referring to the IASLC new grading system, patients were divided into a low/intermediate-grade group and a high-grade group. Radiomic features were selected by using the least absolute shrinkage and selection operator (LASSO), the logistic regression (LR) classifier was used to establish radiomics model (RM), clinical-radiological features model (CRM) and combined rad-score with radiological features model (CRRM), and visualized CRRM by nomogram. The area under the curve (AUC) of the receiver operating characteristic (ROC) curve and calibration curve were used to evaluate the performance and fitness of models.
In the training set, RM, CRM, and CRRM achieved AUCs of 0.825 [95% CI (0.735-0.916)], 0.849 [95% CI (0.772-0.925)], and 0.888 [95% CI (0.819-0.957)], respectively. For the validation set, the AUCs were 0.879 [95% CI (0.734-1.000)], 0.888 [95% CI (0.794-0.982)], and 0.922 [95% CI (0.835-1.000)], and for the test set, the AUCs were 0.814 [95% CI (0.674-0.954)], 0.849 [95% CI (0.750-0.948)], and 0.860 [95% CI (0.755-0.964)] for RM, CRM, and CRRM, respectively.
All three models performed well in predicting pathological grade, especially the combined model, showing CT radiomics combined with radiological features had the potential to distinguish the pathological grade of early-stage IPA.
基于国际肺癌研究协会（IASLC）新的分级系统，探讨CT影像组学联合放射学特征在预测Ⅰ期浸润性肺腺癌（IPA）病理分级中的价值。
回顾性分析294例Ⅰ期IPA患者的术前CT图像和临床资料（159例为训练集；69例为验证集；66例为测试集）。参照IASLC新分级系统，将患者分为低/中级别组和高级别组。采用最小绝对收缩和选择算子（LASSO）选择影像组学特征，使用逻辑回归（LR）分类器建立影像组学模型（RM）、临床-放射学特征模型（CRM）以及影像组学评分与放射学特征联合模型（CRRM），并通过列线图对CRRM进行可视化。采用受试者操作特征（ROC）曲线的曲线下面积（AUC）和校准曲线评估模型的性能和拟合度。
在训练集中，RM、CRM和CRRM的AUC分别为0.825 [95%可信区间（CI）（0.735 - 0.916）]、0.849 [95% CI（0.772 - 0.925）]和0.888 [95% CI（0.819 - 0.957）]。在验证集中，AUC分别为0.879 [95% CI（0.734 - 1.000）]、0.888 [95% CI（0.794 - 0.982）]和0.922 [95% CI（0.835 - 1.000）]；在测试集中，RM、CRM和CRRM的AUC分别为0.814 [95% CI（0.674 - 0.954）]、0.849 [95% CI（0.750 - 0.948）]和0.860 [95% CI（0.755 - 0.964）]。
三种模型在预测病理分级方面均表现良好，尤其是联合模型，表明CT影像组学联合放射学特征有潜力区分早期IPA的病理分级。
