影像组学特征对肺纯磨玻璃结节侵袭性腺癌与 非侵袭性腺癌的鉴别价值

目的研究影像组学特征对肺纯磨玻璃结节（pGGN）侵袭性腺癌与非侵袭性腺癌的鉴别价值。方法回顾性分析2011年7月-2016年7月间156例经手术病理证实为肺腺癌且存在pGGN的病人资料，其中男65例，女91例，年龄37~81岁（中位年龄56岁）。经手术病理证实，非侵袭性腺癌60例（包括非典型腺瘤样增生28例，原位癌32例），侵袭性腺癌96例（包括微浸润腺癌53例,浸润性腺癌43例）。应用图像分析软件ImageJ1.50b提取4大类共68个影像组学特征，采用线性回归对所有影像组学特征进行共线性诊断，将不存在共线性的影像组学特征（54个）作为独立参数来预测pGGN的病理侵袭性。采用二元logistic回归分析建立影像组学特征与pGGN病理类型之间的预测模型，采用向后步进方法选取最佳定量特征，当定量特征P<0.05时纳入模型，当定量特征P>0.10时剔除模型。采用受试者操作特征（ROC）曲线对模型进行分析，并评价影像组学特征预测pGGN病理侵袭性的效能。结果二元logistic回归模型从54个影像组学特征中筛选出8个具有统计学意义的影像组学特征（P<0.05），其中描述肿瘤大小的特征2个（面积和周长），描述肿瘤形态的特征2个（椭圆长轴和椭圆短轴），描述肿瘤灰度直方图的特征3个（众数、最大灰度值和直方图峰度），描述肿瘤纹理的特征1个（灰度共生矩阵熵值）。基于此模型建立的ROC曲线分析显示，曲线下面积（AUC）=0.951（95%CI：0.918~0.985），诊断的敏感度和特异度分别为94.8%和86.7%。结论影像组学特征对鉴别肺pGGN侵袭性腺癌与非侵袭性腺癌有较高的价值，并具有良好的诊断效能。

Radiomic features analysis of pulmonary pure ground-glass nodules for distinguishing invasive adenocarcinoma from non-invasive adenocarcinoma

【Abstract】 Objective To investigate the diagnostic values of radiomic features of patients with pulmonary ground-glass nodule (GGN) associated with pathological invasion. Methods Retrospectively evaluated 156 resected pulmonary GGN radiomic features and pathologically classified them as pre-invasive lesions (AAH, AIS) and invasive lesions (MIA, I-ADC). 68 radiomic features were extracted from segmented of each tumor, and 54 of these, which were considered independent features, were included in the logistic analysis.Receiver operating characteristic(ROC) curve analysis were performed to evaluated the model efficiency. Results There were 60 pre-invasive lesions and 96 invasive lesions among the 156 GGN. 8 radiomic features were independent features significantly associated with GGN pathological invasion（P<0.05）.logistic regression model showed that radiomic features to a significant predicting power(area under the curve [AUC], 0.951).The sensitivity 94.8% and specificity 86.7%. Conclusion radiomic features has a high performance in predicting the invasive of GGN.