Huang Rikun, Zhao Chunli, Yang Jinhan, Lu Bingfeng, Dai Yi, Lin Miaomiao, Zhao Xiang, Huang Haipeng, Pan Xiaoyu, Lu Liling, Chen Lina, Li Kai
Guiping People's Hospital, Guiping, China.
People's Hospital of Guangxi Zhuang Autonomous Region, Nangning, China.
Eur J Radiol. 2025 Jun 12;190:112227. doi: 10.1016/j.ejrad.2025.112227.
OBJECTIVE: To explore the value of a nomogram based on radiomics and computed tomography (CT) features for preoperative prediction of visceral pleural invasion (VPI) of subpleural, small (≤2 cm) invasive adenocarcinoma (IAC) of the lung. METHODS: For this retrospective study, 457 cases of invasive lung adenocarcinoma ≤ 2 cm were collected from three tertiary hospitals in Guangxi and used in a training group (n = 254), validation group (n = 112), and test group (n = 91). Risk factors for IAC VPI were screened by univariate and multivariate logistic regression analyses, and a CT model was constructed. Radiomics features of regions representing the gross tumor area (GTA), peritumor area (PTA), and gross peritumor area (GPTA) were extracted from CT images, and the optimal feature subsets based on radiomics score were selected to construct three radiomics models. A combination model was then constructed from the radiomics model with the optimal radiomics score and the CT model and visualized by nomogram. Model performance was analyzed by receiver operating characteristic curve analysis and DeLong test. RESULTS: Pleural indentation (P < 0.05), pleural thickening (P < 1e-04), and tumor diameter (P < 0.001) were identified as risk factors of the CT model for predicting VPI of IAC. Among 1226 radiomics features, 5, 13, and 12 optimal features were selected for the GTA, PTA, and GPTA models, respectively, and the area under the curve (AUC) values did not differ among these models. Based on AUC values, the CT model and GPTA model features were combined to construct the predictive nomogram. Compared with the individual models, the nomogram exhibited better accuracy, specificity, and AUC values (AUC values for training, verification, and test groups were 0.86, 0.84, and 0.86, respectively). Calibration curve and decision curve analyses showed that the nomogram outperformed traditional CT features and radiomics studies, and could offer greater clinical benefit. CONCLUSIONS: The developed nomogram combining CT and radiomics features shows high diagnostic value for VPI prediction of IAC of the lung.
目的：探讨基于影像组学和计算机断层扫描（CT）特征的列线图对肺亚胸膜下小（≤2 cm）浸润性腺癌（IAC）术前预测脏层胸膜侵犯（VPI）的价值。 方法：在这项回顾性研究中，从广西的三家三级医院收集了457例直径≤2 cm的浸润性肺腺癌病例，并将其分为训练组（n = 254）、验证组（n = 112）和测试组（n = 91）。通过单因素和多因素逻辑回归分析筛选IAC VPI的危险因素，并构建CT模型。从CT图像中提取代表肿瘤大体面积（GTA）、肿瘤周围区域（PTA）和肿瘤大体周围区域（GPTA）的区域的影像组学特征，并基于影像组学评分选择最佳特征子集来构建三个影像组学模型。然后将具有最佳影像组学评分的影像组学模型与CT模型相结合构建组合模型，并通过列线图进行可视化。通过受试者工作特征曲线分析和德龙检验分析模型性能。 结果：胸膜凹陷（P < 0.05）、胸膜增厚（P < 1e - 04）和肿瘤直径（P < 0.001）被确定为CT模型预测IAC VPI的危险因素。在1226个影像组学特征中，分别为GTA、PTA和GPTA模型选择了5个、13个和12个最佳特征，这些模型的曲线下面积（AUC）值无差异。基于AUC值，将CT模型和GPTA模型的特征相结合构建预测列线图。与单个模型相比，列线图表现出更好的准确性、特异性和AUC值（训练组、验证组和测试组的AUC值分别为0.86、0.84和0.86）。校准曲线和决策曲线分析表明，列线图优于传统CT特征和影像组学研究，并且可以提供更大的临床益处。 结论：所开发的结合CT和影像组学特征的列线图对肺IAC的VPI预测具有较高的诊断价值。
Abdom Radiol (NY). 2025-1-12
Zotero 插件