Brain Functional Connectivity Biomarkers Identification Method Based On Neighborhood Decision Rough Set

Brain functional connectivity biomarkers can provide physiological basis for the prediction and diagnosis of neuropsychiatric diseases.Identifying brain functional connectivity biomarkers is an important topic in brain science research.Because the brain functional connectivity data has the characteristics of continuous,high noise,high dimension and small samples,the accurate identification of brain functional connectivity biomarkers is facing great challenges.As a rough set model that can reduce continuous data containing noise,neighborhood decision rough set is expected to provide a new idea for identification of brain functional connectivity biomarkers.However,there are two important problems to be considered:(1)The time complexity of neighborhood decision rough set feature reduction method is relatively high,when facing high-dimensional brain functional connectivity data,the problem of low time performance is particularly serious.Therefore,how to use neighborhood decision rough set to quickly identify brain functional connectivity biomarkers is an important research problem.(2)Brain functional connectivity data have the characteristics of small samples,while the neighborhood decision rough set feature reduction method usually realizes feature reduction based on single granularity,so it can not make full use of sample information to identify brain functional connectivity biomarkers with stronger classification ability.Therefore,how to use neighborhood decision rough set to mine more information from small samples of brain functional connectivity data,and then identify brain functional connectivity biomarkers with stronger classification ability is also a key research problem.In view of the above two problems,this thesis has carried out the following two works:Firstly,a neighborhood decision rough set brain functional connectivity biomarkers identification method based on hash mapping and feature grouping is proposed.This method first maps the subjects into the hash table with array as the storage structure,so that the neighborhood range of any subject is reduced to the array where the subject is located and its adjacent array.On this basis,combined with the symmetry of neighborhood relationship,the neighborhood of the subject is quickly generated;then the brain functional connectivity features are grouped based on the feature separability measure,so as to obtain the coarse-grained search space of the brain functional connectivity biomarkers.The experimental results on ABIDE I and ADNI data sets show that this method can greatly improve the efficiency of using neighborhood decision rough set to identify brain functional connectivity biomarkers,and can obtain brain functional connectivity biomarkers with stronger classification ability compared with multiple brain functional connectivity biomarkers identification method.Secondly,based on the previous work,a neighborhood decision rough set brain functional connectivity biomarkers identification method integrating multigranularity information is proposed.This method uses k-nearest neighbor neighborhood to optimize δ-neighborhood in neighborhood decision rough set,and the new neighborhood optimized by combining a single k value with the corresponding k-nearest neighbor neighborhood is regarded as a single granularity,and then the brain functional connectivity biomarkers are identified by generating and fusing multiple granularity information.Specifically,The method first determines a better performing reference granularity,then generates the granularity under multiple different k values based on the similarity measure,and finally the multi-granularity information is fused with addition-deletion method,so as to identify brain functional connectivity biomarkers.The experimental results on ABIDE I and ADNI data sets show that compared with multiple brain functional connectivity biomarkers identification methods,this method can find brain functional connectivity biomarkers with stronger classification ability after fusing multi-granularity information.In conclusion,according to the characteristics of brain functional connectivity data,this thesis creatively proposes two effective brain functional connectivity biomarkers identification methods based on neighborhood decision rough set.This thesis not only enriches the theory and method system of neighborhood decision rough set,but also provides a new means for the identification of brain functional connectivity biomarkers.Therefore,it has important theoretical significance and application value.