Data Sheet 1_CT-based habitat radiomics for preoperative differentiation of adenocarcinoma in situ/minimally invasive adenocarcinoma from invasive adenocarcinoma manifesting as ground-glass nodules: a multicenter study.docx

ObjectivesTo develop a CT-based habitat radiomics model for preoperative differentiation of adenocarcinoma in situ/minimally invasive adenocarcinoma (AIS/MIA) from invasive adenocarcinoma (IAC) manifesting as ground-glass nodules (GGNs), and to construct a combined model integrating clinical risk factors for optimizing individualized treatment decisions.
MethodsWe retrospectively collected imaging and clinical data from 630 patients with pathologically confirmed ground-glass nodules (GGNs) who underwent surgical resection at two medical centers between January 2020 and December 2024. Patients from Center 1 were randomly divided into training and internal validation sets at a 7:3 ratio, while patients from Center 2 served as the external validation set. Tumor habitats were generated using K-means clustering, and radiomics features were extracted from intratumoral, peritumoral 1mm, peritumoral 2mm, and habitat regions. Feature selection was performed using Least Absolute Shrinkage and Selection Operator (LASSO) regression, and predictive models were constructed using multiple machine learning algorithms. A combined nomogram was developed by integrating the Habitat model, Intratumoral model, and Clinic model. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA).
ResultsIn the training set, the Combined model demonstrated optimal performance (AUC = 0.928), followed by the Habitat model (AUC = 0.924), both significantly outperforming the Intratumoral model (AUC = 0.879), Peritumoral 1mm model (AUC = 0.874), Peritumoral 2mm model (AUC = 0.868), and Clinic model (AUC = 0.807) (P<0.05). In the external validation set, the Combined model maintained superior performance (AUC = 0.897), significantly exceeding all other models (P<0.05). The Habitat model showed the second-best performance in external validation (AUC = 0.840). Hosmer-Lemeshow test and calibration curves demonstrated good calibration for both the Combined and Habitat models across all cohorts. DCA indicated high net benefit for both models in clinical applications.
ConclusionCT-based habitat radiomics effectively quantifies intratumoral heterogeneity, significantly improving the differentiation between AIS/MIA and IAC. The combined nomogram integrating habitat features, intratumoral features, and clinical factors demonstrates excellent diagnostic performance and generalizability, providing a reliable preoperative assessment tool for individualized treatment decision-making in ground-glass nodular lung adenocarcinoma.

研究目的：旨在构建基于CT的肿瘤栖息地影像组学（habitat radiomics）模型，用于术前鉴别以磨玻璃结节（ground-glass nodules, GGNs）为影像学表现的浸润性腺癌（invasive adenocarcinoma, IAC）与原位腺癌/微浸润腺癌（adenocarcinoma in situ/minimally invasive adenocarcinoma, AIS/MIA）；同时构建整合临床风险因素的联合模型，以优化个体化治疗决策方案。
研究方法：本研究回顾性收集了2020年1月至2024年12月期间，于两家医学中心接受手术切除且经病理证实为磨玻璃结节（GGNs）的630例患者的影像与临床资料。其中，中心1的患者以7:3的比例随机划分为训练集与内部验证集，中心2的患者则作为外部验证集。采用K-means聚类生成肿瘤栖息地区域，并从瘤内、瘤周1mm、瘤周2mm及栖息地区域中提取影像组学特征。通过最小绝对收缩和选择算子（Least Absolute Shrinkage and Selection Operator, LASSO）回归进行特征筛选，并采用多种机器学习算法构建预测模型。本研究整合栖息地模型、瘤内模型与临床模型，构建了联合列线图（nomogram）。采用受试者工作特征（receiver operating characteristic, ROC）曲线、校准曲线与决策曲线分析（decision curve analysis, DCA）评估模型性能。
研究结果：在训练集中，联合模型展现出最优性能（曲线下面积AUC=0.928），其次为栖息地模型（AUC=0.924），二者性能均显著优于瘤内模型（AUC=0.879）、瘤周1mm模型（AUC=0.874）、瘤周2mm模型（AUC=0.868）与临床模型（AUC=0.807）（P<0.05）。在外部验证集中，联合模型仍保持优异性能（AUC=0.897），显著优于其余所有模型（P<0.05）；栖息地模型在外部验证集中表现次之（AUC=0.840）。Hosmer-Lemeshow检验与校准曲线结果显示，联合模型与栖息地模型在所有队列中均具有良好的校准度。决策曲线分析结果表明，两种模型在临床应用中均具有较高的净获益。
研究结论：基于CT的肿瘤栖息地影像组学可有效量化瘤内异质性，显著提升AIS/MIA与IAC的鉴别效能。整合栖息地特征、瘤内特征与临床因素的联合列线图具备优异的诊断性能与泛化能力，可为磨玻璃结节型肺腺癌的个体化治疗决策提供可靠的术前评估工具。