放射组学特征联合CT影像学征象对周围型肺癌胸膜浸润的预测价值

Prognostic value of radiomics features combined with CT imaging signs in pleural invasion of peripheral non-small cell lung cancer

目的探讨放射组学特征与CT影像学征象相结合在预测周围型非小细胞肺癌(NSCLC)发生胸膜浸润(VPI)中的价值。方法选取经手术病理证实的周围型NSCLC患者398例,根据有无胸膜浸润将患者分为阴性组209例和阳性组189例,评估所有患者的CT征象。以7∶3的比例将患者随机分配到训练集和验证集,使用随机森林回归分析构建预测模型(纹理特征模型、CT影像学特征模型和联合预测模型),用ROC曲线评价其诊断性能。结果阴性组和阳性组患者在平均直径、平均CT值、密度、病灶与胸膜的关系(RAP)分型、胸膜凹陷征方面差异均有统计学意义(P<0.05),且多因素回归分析显示肿瘤平均直径、密度、RAP类型和淋巴结转移是VPI的独立预测因子。共选出786个纹理参数,通过mRMR和LASSO特征分析识别出12个具有预测意义的纹理特征。RF回归分析构建预测模型,结果显示联合预测模型对训练集和验证集的AUC分别为0.915、0.887,且联合预测模型的AUC分别显著高于纹理特征模型和CT影像学特征模型(0.915 vs 0.856 vs 0.852;0.887 vs 0.855 vs 0.827);联合预测模型的特异度也高于纹理特征模型和CT影像学特征模型(91.2%vs 88.2%vs 85.3%;94.1%vs 85.3%vs 70.6%)。结论放射组学特征联合CT影像学征象可有效预测直径≤3.0 cm周围型NSCLC患者胸膜浸润的存在。

Objective To investigate the value of combining radiomic features with CT imaging findings in the prediction of pleural invasion in peripheral non-small cell lung cancer.Methods Clinical and imaging data of 398 patients with peripheral NSCLC confirmed by surgery and pathology were retrospectively collected.The patients were divided into VPI(-)group(209 cases)and VPI(+)group(189 cases)according to the pathological findings of pleural infiltration.CT signs of all patients were evaluated.The patients were randomly assigned to the training set and validation set in a ratio of 7:3.The prediction models,in terms of texture feature model,CT imaging feature model and joint prediction model,were constructed by random forest regres‐sion analysis,and the diagnostic performance was evaluated by ROC curve.Results Comparative analysis of CT imaging signs showed that there were statistically significant differences in mean diameter,mean CT value,density,focus-pleural relationship(RAP)typing and pleural depression sign between the two groups(P<0.05),and multivariate regression analysis showed that mean tumor diameter,density,RAP type and lymph node metastasis were independent predictors of VPI.A total of 786 texture parameters were selected,and 12 texture features with predictive significance were identified through mRMR and LASSO feature analysis.RF regression analysis constructed predictive models.The AUC of the combined prediction model was significantly higher than that of the texture feature model and CT imaging feature model(0.915 vs 0.856 vs 0.852;0.887 vs 0.855 vs 0.827).Conclusion Radiomic features combined with CT imaging findings can effectively predict the presence of pleural invasion in patients with peripheral non-small cell lung cancer≤3.0 cm in diameter.