Research On Feature Extraction And Selection Methods Of The Brain Functional Network

Brain function networks based on neuroimaging data have become an important tool for understanding brain functional connections in patients with mental disorders.The network is constructed by analyzing the activity patterns between different regions of the brain,which helps to understand how different regions of the brain work together to perform various cognitive and behavioral tasks.However,due to its complexity and variability,as well as the lack of standardized network analysis methods,it is difficult to explain the results of research on mental disorders based on brain functional networks from the perspective of disease mechanisms.At present,extracting and selecting key features from brain functional networks to identify abnormal connection patterns in neurological diseases is a promising method.Therefore,this paper has carried out the following research on the construction method of brain functional networks and feature extraction and selection methods from the perspective of machine learning:(1)A method for constructing high-order brain functional network based on electroencephalography is proposed.This method first generates brain network sequences using the sliding window technology and correlation algorithm for all channels in the electroencephalogram,which is called the low-order brain functional network.Then,the functional connectivity time series between each channel and other channels in the low-order brain functional network are clustered into several clusters by clustering algorithm,and the correlation between clusters is calculated to estimate the high-order brain functional network.This method has the following advantages: 1)It captures the complex interaction relationships between multiple channels,which can further understand and study the information processing mechanisms of the brain,thereby better simulate the deep mechanisms of the brain;2)The high-order brain functional network constructed based on the clustering algorithm can avoid the high-dimensional complex problems caused by the direct use of low-order brain functional network to construct high-order brain functional network;3)The fusion of low-order and high-order brain functional networks can effectively improve the reliability and accuracy of the brain functional network,and more comprehensively assist in the diagnosis and analysis of patients with depression.(2)A method for extracting the state transition feature based on coarse-grained similarity measurements is proposed.This method first generates the brain functional network using the sliding window technology and correlation algorithm for all the regions of interest in resting-state functional magnetic resonance imaging,and then extracts the seven central invariant moment features of the brain functional network based on the coarse-grained similarity measurement method.Further,all seven central invariant moment features extracted are divided into multiple clusters,and each cluster represents a state.Finally,the features of the state itself and the transition features among multiple states are extracted for analysis.This method has the following advantages: 1)The state division based on the coarse-grained similarity measurement method can effectively avoid the problems of high computational complexity and deviation of state division caused by the high-dimensions data;2)Considering the transition features among multiple states,which effectively capture the complex and subtle information of the brain and provide a new perspective for understanding the neurophysiological mechanisms of autistic patients;3)The fusion of the features of the state itself and the transition features among multiple states can obtain a more comprehensive state description,improve the accuracy and reliability of state recognition,and provide more abundant identification information for the diagnosis and analysis of autistic patients.