Research On Neighborhood Rough Set With Fish Swarm Algorithm For Feature Reduction In FMRI Brain Functional Connectivity

Brain functional connectivity depicts the dynamic correlation of neuronal activities between different brain regions,which provides a new perspective for people to understand the pathological mechanism of brain diseases.In recent years,the classification of brain functional connectivity based on f MRI has attracted extensive attention of researchers because it can find important features of brain functional connectivity related to a certain brain disease,which is of great significance for the early diagnosis and treatment of brain diseases.However,the high-dimensional smallsample-size and multi-noise of f MRI brain functional connectivity data pose a huge challenge to the construction of classification models.In order to deal with this challenge effectively,the feature reduction of f MRI brain functional connectivity has become a key research topic in the classification of brain functional connectivity.However,most of the existing feature reduction methods in f MRI brain functional connectivity lack of in-depth analysis and effective use of f MRI data.Rough set is a mathematical theory dealing with imprecise,inconsistent and incomplete knowledge.It can obtain the core knowledge of data without providing any prior information and realize feature reduction.Since f MRI brain functional connectivity is continuous data,this paper introduces the neighborhood rough set theory which can directly process continuous data into the research of feature reduction for brain functional connectivity,and carries out the following two research work around the steps of feature subset evaluation and feature subset generation in this research:(1)In order to improve the effectiveness of searching feature subset,we propose an algorithm based on algebraic neighborhood rough set with fish swarm algorithm for feature reduction in f MRI brain functional connectivity(ANRS-FSA).This method evaluates the correlation between feature subsets and tags based on the dependence degree in the algebraic neighborhood rough set theory,and uses fish swarm algorithm with global optimization to search the feature subsets with strong discrimination ability.Specifically,this method initializes and evaluates the fish individuals according to the dependency information of the feature subset;and then the preying,swarming,following mechanisms of the fish swarm algorithm,as well as two new simulation mechanisms of crossover and migration are used to continuously search the feature subset of brain functional connectivity with strong discrimination ability.The experimental results on three f MRI datasets of brain diseases show that the new method can not only effectively reduce the dimension of brain functional connectivity,but also obtain the features of brain functional connectivity with stronger classification and discrimination ability compared with many excellent feature reduction methods.(2)In order to improve the efficiency of fish swarm algorithm to search for brain functional connectivity feature subset,from the perspective of informational neighborhood rough set,we propose an algorithm based on informational neighborhood rough set with double-population fish swarm algorithm for feature reduction in f MRI brain functional connectivity(INRS-DFSA).This method evaluates the correlation between feature subsets and tags based on the neighborhood mutual information in the informational neighborhood rough set theory,and uses the double-population fish swarm algorithm to search the feature subsets with strong discrimination ability.Specifically,this method determines the primary selection feature set based on the neighborhood mutual information of the feature,then initializes the fish individuals on the primary selection set,and evaluates the individual based on the neighborhood mutual information of the feature subset;Finally,we implement the strategy of doublefish population,divide the fish population into an elite-fish population and a commonfish population,use the elite-fish population’s preying and swallowing mechanisms and the common-fish population’s swarming,following and swallowing mechanisms to continuously search the feature subset of brain functional connectivity with strong discrimination ability.The experimental results on three f MRI datasets of brain diseases show that INRS-DFSA algorithm can achieve better performance than the previous one,and has obvious advantages compared with other feature reduction algorithms.