基于影像组学鉴别不同病理类型的肺癌脑转移

Differentiating Brain Metastases from Different Pathological Types of Lung Cancers Using Radiomic

目的探讨基于影像组学结合MRI的影像组学特征鉴别不同病理类型的肺癌脑转移。方法回顾性分析69例经病理诊断为非小细胞肺癌脑转移(NSCLC METs)和小细胞肺癌脑转移(SCLC METs)的患者MRI影像资料。使用不同序列的组合,从T1WI,T2WI和对比增强T1WI中提取每例病例的影像组学特征。通过Variance Threshold, SelectKBest和Lasso方法对影像组学特征进行筛选后,建立了包括随机森林(RF),K邻近(KNN)、支持向量机(SVM)、逻辑回归(LR)和决策树(DT)在内的5个分类器和35个影像组学模型。结果提取影像组学特征(n=4227),选择了30个特征构建影像组学模型。使用KNN分类方法与T1WI、T2WI和对比增强T1WI序列图像相结合而构建的影像组学模型获得了最佳鉴别效果,NSCLC METs的曲线下面积(AUC)0.94,敏感度0.88,特异度0.87,SCLC METs的AUC 0.94,敏感度0.87、特异度0.88。结论使用KNN分类方法与T1WI、T2WI和对比增强T1WI序列图像相结合而构建的影像组学模型对鉴别肺癌脑转移具有一定的价值。

Objective To differentiate non-small cell lung cancer brain metastases(NSCLCMETs)from small cell lung cancer brain metastases(SCLCMETs)by using radiomic features derived from Magnetic resonance imaging(MRI) and multiple classifiers. Methods Sixty-nine patients who were pathologically diagnosed with non-small cell lung cancer brain metastases and small cell lung cancer brain metastases were retrospectively analyzed.Radiomic features of each case were extracted from T1 weighted imaging, T2 weighted imaging and contrast-enhanced T1-weighted imaging.We used VarianceThreshold, SelectKBest and Lasso classification method to select features.Fivemachine learning classification methods, including random forests(RF),k-nearest neighbor(KNN),support vector machine(SVM),Logistic regression(LR) and decision tree(DT) algorithm were then trained using the 5-fold cross validation strategy to distinguish brain metastases with non-small cell lung cancer and small cell lung cancer.Furthermore, we studied the combined use of different sequences. Results A total of radiomic(n=4227) features were extracted and 30 features were selected for each case to build radiomics model. The model built by combining different sequences that used KNN classification method obtained the greatest performance with AUC were 0.94,sensitivity=0.88,specificity=0.87 in NSCLC METs, and AUC were 0.94,sensitivity=0.87,specificity=0.88 in SCLC METs, respectively. Conclusion The combination of multiple sequences and the use of KNN machine classification method have certain value for identifying different lung cancer brain metastases.