双能CT生境放射组学术前评估肺磨玻璃结节浸润性
本研究回顾性纳入199例患者(267个GGNs)作为内部队列、52例患者(54个GGNs)作为外部队列,均为组织学证实的微浸润腺癌或浸润性腺癌。基于双层光谱探测器CT重建电子密度图像,通过k-means聚类划分生境亚区域并提取放射组学特征,构建XGBoost模型。融合ED与常规CT亚区域特征的生境模型表现最佳,内部测试集AUC为0.9052、外部测试集AUC为0.8414,可用于术前区分IAC与MIA。
要点
- 融合ED与常规CT生境亚区域特征的模型在内部测试集AUC达0.9052,外部测试集为0.8414,对GGNs浸润性术前判断具有参考价值。
- 非生境模型中ED图像方法优于常规CT(内部测试集AUC 0.7491 vs 0.7303,外部测试集0.7366 vs 0.7090)。
- 生境亚区域通过k-means聚类基于体素强度值划分,提示结节内不同亚区域的放射组学特征对鉴别IAC与MIA有增量价值。
核心结论
- 基于双能CT的生境放射组学策略为肺GGNs术前区分IAC与MIA提供了一种有前景的方法,可支持个体化治疗规划。
- 融合电子密度图像与常规CT图像亚区域特征的生境模型取得了最佳的预测性能。
研究设计与样本
- 回顾性研究,内部队列199例患者(267个GGNs),外部队列52例患者(54个GGNs),均为组织学证实的MIA或IAC。
- 所有患者均采用双层光谱探测器系统行常规CT成像,从光谱基图像重建获得电子密度图像。
- 内部队列按7:3随机分为训练集和测试集,采用XGBoost构建模型,以AUC及95%CI评估模型性能。
关键影像表现/征象
- 生境亚区域通过对体素强度值进行k-means聚类划分,从各亚区域提取放射组学特征用于模型构建。
诊断标准与关键数值
- 非生境模型:ED图像方法在内部测试集AUC为0.7491(95%CI,0.5930-0.8799),常规CT为0.7303(95%CI,0.6005-0.8514)。
- 非生境模型:ED图像方法在外部测试集AUC为0.7366(95%CI,0.5986-0.8586),常规CT为0.7090(95%CI,0.5556-0.8436)。
- 生境模型:融合ED与常规CT亚区域特征的模型内部测试集AUC为0.9052(95%CI,0.8255-0.9676),外部测试集AUC为0.8414(95%CI,0.7215-0.9361)。
临床应用与阅片要点
- 日常阅片中对表现为磨玻璃结节的肺腺癌,可关注结节内部不同亚区域的影像特征,而非仅依赖整体结节特征进行浸润性判断。
- 电子密度图像在非生境模型中表现优于常规CT,提示双能CT重建的ED图像对GGNs浸润性评估具有潜在增量价值。
- 生境分析思路提示结节内部异质性对鉴别IAC与MIA有重要意义,报告时注意描述结节内部密度或信号的不均质性。
局限性
- 本研究为回顾性设计,内部队列与外部队列样本量有限。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
RATIONALE AND OBJECTIVES: This study aimed to preoperatively predict the invasiveness of ground-glass nodules (GGNs) using habitat-based radiomics derived from dual-energy computed tomography (CT).
MATERIALS AND METHODS: This study retrospectively included 199 patients (267 GGNs) in the internal cohort and 52 patients (54 GGNs) in the external cohort, all with histologically confirmed minimally invasive adenocarcinoma (MIA) or invasive adenocarcinoma (IAC). All patients underwent conventional CT imaging using a dual-layer spectral detector system, and electron density (ED) images were reconstructed from the spectral base images. Habitat subregions were delineated by applying k-means clustering to voxel intensity values, and radiomic features were subsequently extracted from these subregions. The internal cohort was randomly divided into training and test sets in a 7:3 ratio. Extreme gradient boosting (XGBoost) models were developed, and model performance was evaluated using area under the curve (AUC) with 95% confidence intervals (CIs).
RESULTS: For nonhabitat models, the ED-based approach outperformed conventional CT in both the internal test set (AUC: 0.7491; 95% CI, 0.5930-0.8799 vs. 0.7303; 95% CI, 0.6005-0.8514) and the external test set (AUC 0.7366; 95% CI, 0.5986-0.8586 vs. 0.7090; 95% CI, 0.5556-0.8436). For habitat-based models, the integration of subregional features from both ED and conventional CT images achieved the best predictive performance, yielding an AUC of 0.9052 (95% CI, 0.8255-0.9676) in the internal test set and 0.8414 (95% CI, 0.7215-0.9361) in the external test set.
CONCLUSION: This habitat-based radiomics strategy may offer a promising approach for preoperative differentiation of IAC from MIA in pulmonary GGNs, supporting personalized treatment planning.
引用
- Yang, Kang, Lv, et al. Habitat-Based Radiomics on Dual-Energy CT for Preoperative Assessment of Invasiveness in Lung Adenocarcinoma with Ground-Glass Nodules. Acad Radiol. 2026. PMID: 41565493. doi:10.1016/j.acra.2026.01.004
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。