Brain Functional Zoning Based On Improved BIRCH Algorithm For Rs-fMRI Data Of AD Patients

Alzheimer's Disease(AD)ranks first in the burden of social diseases due to its insidious onset and scarce treatment.Understanding the state of brain region in AD patients is very important for understanding the pathogenesis of the disease and preventing the development of the disease.At present,the commonly used AAL partition and BA partition are divided according to the anatomical structure of cells.The lack of functional specificity in the analysis of brain function by structural zoning may lead to some errors and even wrong conclusions and brain function partition can avoid above problem.Rs-fMRI can reveal the spontaneous activity pattern and connection pattern of the brain,which is the main imaging method to study the state of AD brain region.Functional division of brain by hierarchy is essentially a hierarchical clustering problem of brain voxel,BIRCH algorithm is a kind of agglomerated hierarchical clustering method based on distance,which is suitable for rs-fMRI data with a large amount of data,and time complexity is low,occupy memory is little.Therefore,This paper proposes a brain function partitioning method based on improved BIRCH algorithm for rs-fMRI data of AD patients.The core of this algorithm is the construction of clustering feature Tree(CF-Tree).The construction of BIRCH algorithm CF-Tree is determined by two important parameters:branch factor B and threshold value T,B determines the height and size of the tree and T controls the boundary of the cluster.Therefore,this paper analyzes the important parameters of the algorithm and proposes a BIRCH parameter determination method based on cophenet correlation coefficient.The cophenet correlation coefficient of hierarchical tree was used as an evaluation index to measure the consistency of clustering results with the actual situation,so as to find a reasonable parameter value suitable for rs-fMRI data clustering.According to the clustering of cluster radius distance in original BIRCH algorithm,the rs-fMRI data with the same function but far apart voxels cannot be clustered into one group.So,in this paper,a rs-fMRI data brain functional partitioning method based on spatial density adjustable distance BIRCH(Spatial Density Adjustable Distance BIRCH,SDAD-BIRCH)is proposed.This method avoids the limitation of only considering the average distance from the sample point to the cluster center as a similarity measure,and takes the adjustable distance of the density set as a new measure in the algorithm.Experimental results show that determining the optimal value of parameters in BIRCH algorithm based on cophenet correlation coefficient can reduce the number of times of reconstruction of CF-Tree,obtain a better clustering effect of CF-Tree,and can view the partition results of rs-fMRI data at different levels.The improved BIRCH algorithm based on the space density adjustable distance was compared with the original BIRCH algorithm,the improved BIRCH based on the sum of dispersion squared,and the improved BIRCH based on the weighted similarity,the cluster evaluation index is improved obviously.It shows that the improved algorithm is suitable for rs-fMRI data clustering of AD patients and can obtain more reasonable brain functional partition.