孤立性肺结节诊断模型的特征选择算法

孤立性肺结节诊断模型中未得到充分解决的一个关键问题就是如何选择合适的特征子集。为了构建一个良好的诊断预测模型，提高肺结节良恶性诊断的效率以及准确率，提出了一种基于联合互信息的混合模型特征子集选择算法。该算法综合过滤式和包裹式特征选择模型各自的优势，首先使用过滤式方法得到与诊断有高相关度的候选特征子集，然后通过包裹式方法对候选特征子集进行特征间冗余分析，最后得到最优特征子集。实验表明，该算法与基于其他互信息的过滤式、混合模型特征选择方法相比，不仅在特征子集数目上，而且在良恶性诊断的敏感性、特异性和平均分类准确率上，均具有很好的性能效果。

Feature selection algorithm for diagnostic model of solitary pulmonary nodules

It is a key problem that how to choose an appropriate feature subset in solitary pulmonary nodules diagnosis model.A hybrid model feature subset selection algorithm is proposed to improve the pulmonary nodules diagnosis efficiency and accuracy of benign or malignant,based on joint mutual information.It takes advantages of both filter and wrapper model.Firstly,a filter method is applied to obtain a candidate subset with high correlation.Then,a wrapper method is used to analyze the redundancy between features of the candidate subset.Finally,an optimal feature subset of solitary pulmonary nodules is achieved.Compared with other filter or hybrid feature selection algorithms based on mutual information,the performance of proposed method is better not only on the number of feature subset,but also on the sensitivity,specificity and the average classification accuracy in solitary pulmonary nodules diagnosis.