基于Relief系列算法的脑网络特征选择与分类  

Feature Selection and Classification of Brain Network Based on Relief Algorithms

作　　者：李国豪 杨豪 刘彦[1] 张俊然[1] LI Guo-hao;YANG Hao;LIU Yan;ZHANG Jun-ran(College of Electrical Information,Sichuan University,Chengdu Sichuan 610065,China;West China Hospital,Sichuan University,Chengdu Sichuan 610065,China)

机构地区：[1]四川大学电气工程学院,四川成都610065 [2]四川大学华西医院,四川成都610065

出　　处：《计算机仿真》2022年第10期354-358,共5页Computer Simulation

摘　　要：针对当前脑网络计算产生的多个脑影像指标无法有效地被筛选从而达到去除冗余特征的目的等问题,采用Relief算法及其衍生的多阶段Relief(MS-Relief)算法与特征加权信息熵Relief(LIE-Relief)算法对影像指标进行特征选择。以糖尿病脑网络影像指标特征选择为例,在不同尺度上与双样本T检验筛选的特征进行结果对比。结果表明采用Relief系列算法能够有效地对脑网络拓扑指标进行筛选,为计算出复杂网络指标后的特征选择和模式分类提供了有效方法。Aiming at the problem that the multiple brain image indicators generated by the current brain network calculations cannot be effectively screened to achieve the purpose of removing redundant features, the Relief algorithm and its derivative, Multi-Stage Relief(MS-Relief) algorithm and Local consistency Information Entropy-Relief algorithm(LIE-Relief),were used for feature selection of image indicators. Feature selection of diabetic brain network image indicators was taken as an example, and the results were compared with those of the two-sample T-test from different scales. Experimental results show that Relief-based algorithms can effectively screen the topological indicators of brain networks and provide an effective method for feature selection and pattern classification after calculating complex network indicators.

关 键 词：复杂网络 网络拓扑指标 特征选择 糖尿病 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]