一种用于影像组学的多级特征选择方法

针对医学影像中高维特征的问题,提出一种用于影像组学的多级特征选择方法(MSOM-GA)。用组内相关系数过滤对边界敏感的特征。用混合F-Score和信息增益的方法,去除不相关特征。用遗传算法去除冗余特征,选择最优特征子集。该算法在河南省人民医院脑胶质瘤影像数据上进行验证,实验显示,特征选择算法能显著提升特征的质量,算法的auc、acc、敏感度、特异性分别为0.9756、92.29%、93.70%、89.26%,与对照组相比具体数值有较为显著的提升。结果表明,该特征选择方法能有效地去除肿瘤边界不稳定特征、冗余特征和不相关特征,提高模型的训练精度。

A MULTI-LEVEL FEATURE SELECTION METHOD FOR RADIOMICS

To solve the problem of high dimensional features in medical images,a multi-level feature selection method(MSOM-GA)for Radiomics is proposed in this paper.The boundary sensitive features were filtered by the intra-group correlation coefficient.Then a method combining F-Score and information gain was used to remove uncorrelated features.The genetic algorithm was used to remove redundant features and select the optimal feature subset.It was verified on the glioma image data of Henan Provincial People s Hospital.The experimental results show that the feature selection algorithm can significantly improve the quality of features.The auc,acc,sensitivity and specificity of the proposed algorithm were 0.9756,92.29%,93.7%and 89.26%,respectively.Compared with the control group,the specific values were significantly improved.The results show that the proposed feature selection method can effectively remove the unstable,redundant and irrelevant features of the tumor boundary,and it improves the training accuracy.