Multifunctional Clustering Of Cerebral White Matter Fibers

As the most advanced part of the nervous system,the human brain has a very complex function and structure.A normal operation of brain functions requires the white matter fibers to finish the transmission and exchange of signals between neurons,which makes the brain white matter fibers plays an extremely important role in the brain function network.Combined with the known clustering methods of white matter fibers,it is found that these methods only consider the functional role of white matter fibers in the process of white matter fiber clustering,whether from the perspective of white matter fiber geometry,anatomical map or fiber functional characteristics.Under the background of Neurology,although the function of a bundle of white matter fibers is consistent,the same white matter fiber may participate in different functional activities.In this article,we used Diffusion Tensor Imaging(DTI)data to provide the direction and spatial distribution of the main white matter fibers in the brain,combined with the resting-state functional magnetic resonance(rs-fMRI).Based on the functional activity signals of white matter fibers in the data of resonance(rs-fMRI),the multi-functional Clustering Research of white matter fibers in the brain was carried out from the perspective of the temporal characteristics of white matter fibers,aiming to design a multi-functional clustering algorithm of white matter fibers in the brain,which can intuitively show the functional connectivity and different functional participation of white matter fibers in the whole brain.The research contents of this dissertation are as follows:(1)Firstly,Preprocess DTI data and rs-fMRI data,obtain the whole brain white matter fibers from DTI data,and then register the rs-fMRI data into their respective DTI space to extract the white matter fiber time series information data,that is,obtain the resting state functional signal of the whole brain white matter fibers.(2)Secondly,a multifunctional clustering model of the white matter fibers which based on sparse dictionary learning is proposed.Combining sparse representation and dictionary learning,the temporal information of the whole brain resting white matter fiber data is used as the input data to learn the temporal characteristics of the whole brain white matter fiber.Then,the white matter fiber is divided into multi-functional groups,and a multi-functional clustering result of white matter fiber is obtained based on the temporal information of white matter fiber.Finally,the method is compared with the automatic fiber clustering method which is based on Affinity Propagation(AP),it is proved that this method has a deeper research significance on the multi-functional properties of white matter fibers.(3)Thirdly,according to the existing deep learning method,a novel multi-functional clustering algorithm of white matter fiber is proposed,which is a multi-functional clustering model of white matter fiber based on self-attention mechanism and convolutional self-encoder.Taking the time sequence information of the whole brain resting white matter fiber data as the input data,the convolution self-encoder model based on self-attention mechanism is used to learn the more discriminative deep features of the white matter fiber time sequence,extract the features and divide the white matter fiber into multi-functional parts,and obtain the multi-functional clustering results of the whole brain white matter fiber.Finally,the two algorithms proposed in this dissertation are compared and the definition is given an average correlation calculation formula is proposed,and it is concluded that this method has better performance in learning deep features.