Design And Implementation Of Deep Clustering System For Brain Functional Connectivity Data

Resting-state functional magnetic resonance imaging(rs-f MRI)is a powerful neuroimaging technique that can be used to study the intrinsic functions and dysfunctions of the brain.It has received much attention in brain medicine research and is widely used in diagnosing and identifying brain diseases,making it a popular area of current research.Brain functional connectivity data is a type of data that is generated from magnetic resonance imaging.Analyzing the connections between different brain areas by constructing brain functional connectivity networks is a common technique used in diagnosing brain disorders at this stage.In recent years,in many studies,brain functional connectivity data have been used to conduct unsupervised analysis studies.Brain functional connectivity data suffers from dimensional catastrophe and complex structure,which makes traditional clustering methods often fail to achieve the desired clustering performance.Thanks to the powerful feature extraction ability of deep learning,traditional clustering methods are gradually being replaced by deep learning.People extract low-dimensional features of data based on deep learning and then cluster them.Therefore,it is of great importance to combine deep learning and clustering for the study of brain diseases.In this paper,we investigate the brain functional connectivity through deep learning and clustering methods.The main work is as follows:(1)Aiming at the poor clustering effect of brain functional connectivity data with high dimensionality and nonlinear structure,a network framework of dual-stream encoders combined with spectral constraints(DENs-SCC)is proposed.This method is based on the fact that the autoencoder considers two different pieces of information when encoding the input data:(1)information between adjacent nodes,(2)discriminative features,and then the encoded features are passed through the decoder decoding so that low-dimensional features maintain local structure.Furthermore,to improve the clustering effect,we impose constraints on the features of the low-dimensional subspace to guide its clustering optimization.Experimental results show that this method can effectively improve the clustering results of brain functional connectivity data.(2)A self-supervised framework based on a soft orthogonal constrained dual-stream encoder(SSCDE)is proposed to address the problems of insufficient information mining due to small samples of brain functional connectivity data and redundancy of information mined by multiple encoders.This framework can mine more information to improve the clustering performance.The method is divided into a pretext task and a downstream task.The pretext task assists the target domain in generating pseudo-labels through domain adaptive learning,and the downstream task designs a dual-stream encoder to reduce redundancy and avoid negative coding to encode samples.Meanwhile,a network is designed to guide the features in the lowdimensional space for clustering optimization.Further,the model is trained self-supervised using the pseudo-labels given by the pretext task,thus improving the clustering of features.The experimental results show that the method can effectively enhance the clustering results of brain functional connectivity data.(3)A prototype clustering system based on brain functional connectivity data was designed and developed,which mainly includes three functional modules: data settings,data processing,and disease identification,to achieve optimal clustering of brain functional connectivity data under different conditions.The system is capable of training,extracting,and clustering data into various methods of clustering models according to the type and quantity of existing data sets to obtain the clustering results of the data sets and thus achieve the purpose of brain disease identification and analysis.The system can assist doctors and others in their work and reduce the difficulty of current medical research on brain diseases in a limited way.