18F-FDG PET/CT影像组学预测非小细胞肺癌亚型的研究

目的:探究基于治疗前18F-FDG PET/CT影像组学特征预测非小细胞肺癌（NSCLC）病理亚型的可行性。方法:回顾性分析100例NSCLC患者治疗前的18F-FDG PET/CT图像，其中腺癌60例，鳞癌40例。首先在PET图像上勾画大体肿瘤靶区（GTV），从GTV内提取肿瘤代谢参数和纹理参数。使用Pearson相关系数和ROC曲线评估特征预测NSCLC病理亚型的效能，并计算其敏感性、特异性和最佳阈值。结果:共提取107个特征，有87个特征在鳞癌与腺癌之间差异有统计学意义（P<0.05）。其中，有8个特征与病理类型具有相关性（r>0.4），AUC均高于0.7。逆差矩、同质性、短区域因子作为预测因子，其ROC曲线下面积分别达到0.770、0.768和0.754，其敏感性和特异性分别为0.949和0.475、0.795和0.607、0.821和0.639。结论:腺癌、鳞癌的部分影像组学特征反应的肿瘤异质性有望为病理诊断提供一种高效、无创的检测方法。

18F-FDG PET/CT radiomic features for predicting the subtypes of non-small-cell lung cancer

Abstract: Objective To investigate the feasibility of using pretreatment 18F-FDG PET/CT radiomic features to predict the pathological subtypes of non-small-cell lung cancer (NSCLC). Methods The pretreatment 18F -FDG PET/CT images of 100 NSCLC patients, including 60 adenocarcinoma (ADC) patients and 40 squamous cell carcinoma (SqCC) patients, were analyzed retrospectively. After the gross tumor volume was delineated on PET images, the metabolic parameters and texture parameters were extracted from gross tumor volume. Pearson correlation coefficients and receiver operating characteristic curve were used to assess the performances of the predictive features in the prediction of the pathological subtypes of NSCLC, and to calculate the sensitivity, specificity and optimal threshold of these features. Results Of 107 features extracted in this study, 87 features reflected the differences between ADC and SqCC (P<0.05). Among the 87 features, there were 8 features related to the pathological subtypes (r>0.4), and their AUC values were all higher than 0.7. Three features with the best predictive performance, namely inverse difference moment, homogeneity and short-zone emphasis, were selected as predictive factors. The AUC values of the 3 predictive factors reached 0.770, 0.768 and 0.754, respectively, and their sensitivity and specificity were 0.949 and 0.475, 0.795 and 0.607, 0.821 and 0.639, respectively. Conclusion Tumor heterogeneity reflected in the radiomic features of ADC and SqCC is expected to provide an efficient and non-invasive detection method for the diagnosis of tumor subtypes.