Research On Abnormal Pattern Mining Of Schizophrenics Based On Brain Information Transfer Network

The human brain is a complex network system,and the realization of cognitive function depends on the information transfer between brain regions.Information transfer is a process in which information flows interact among brain regions.Therefore,information transfer,as an important responsibility of the brain,is the focus of current research.Schizophrenia is a mental disorder with abnormal cognitive function.Episodic memory impairment is schizophrenics(SCH)universal existence of one of the key characteristics.A large number of literatures have shown that the episodic memory disorder in SCH is mainly caused by abnormal connectivity of functional brain network.Therefore,it is necessary to explore the episodic memory disorder in SCH based on task-state functional brain network.At present,when constructing task-state functional brain networks,traditional methods such as correlation or regression between time series are generally adopted.Although the correlation between brain regions is described to some extent,only the interaction between brain regions can be obtained,and specific information transfer modes cannot be extracted from them.To solve the above problems,this study combined the resting state inherent network with the information transfer mapping algorithm to analyze the transfer value of episodic memory task information in the resting state network and construct the information transfer network.In combination with graph theory,anomaly patterns such as information transfer mode,information transfer intensity and topological attribute of SCH are mined,and a classification model is constructed to classify SCH.The main research contents and results are as follows:(1)Explore the abnormality of information transfer between SCH’s brain.Based on the resting state and episodic memory task state FMRI data of 56 Normal Control subjects(NC)and 45 SCH,the information transfer mapping algorithm was used to mine SCH’s abnormal information transfer patterns and information transfer intensity under episodic memory tasks(encoding tasks and retrieval tasks).The results show that visual network,cingulo-opercular network,default network,frontoparietal network and hippocampal network are the most frequently abnormal modules in coding task.In retrieval task,cingulo-opercular network,default network,and frontoparietal network are the most frequently abnormal modules.The correlation and significance of SCH’s information transfer intensity and memory performance in these modules are significantly different from NC.The research indicates that the SCH’s abnormal information transfer pattern and the intensity of information transfer may be an important cause of the episodic memory disorder in SCH.(2)The topology attributes of the information transfer network under episodic memory task were calculated by using graph theory analysis,and the anomaly patterns of SCH were mined by comparing the topology attributes of NC and SCH.The results show that the characteristic path length and standardized characteristic path length of SCH increase significantly,while the global efficiency and local efficiency decrease significantly.Efficiency of node,the node degrees and node betweenness centrality of significant differences between brain areas are mainly distributed in the visual cortex,insula + frontal cortex,the prefrontal cortex,medial temporal lobe,the anterior cingulate + the medial prefrontal cortex,dorsolateral prefrontal cortex,central paracentral lobule + cingulate cortex,somatosensory + motor cortex,parietal cortex,premotor cortex and the temporoparietal pillow border,posterior cingulate cortex.In addition,compared with NC,the correlation and significance of SCH’s node efficiency and memory ability are significantly different in the above cortex.The results show that topology attribute analysis based on information transfer network can effectively detect the SCH anomalies,and these anomalies are an important cause of SCH episodic memory disorder,which can be used as a feature to assist the diagnosis of SCH.(3)SCH was classified by support vector machine,proximity algorithm,naive Bayes and random forest,respectively,which were characterized by significantly different topological attributes in information transfer network under episodic memory task.It is found that the classification accuracy of the information transfer network is improved by about 10% and 8%respectively,compared with the traditional classification model constructed by the resting state and task state networks.The results show that the information transfer network combining the resting state and task state can provide more effective and comprehensive feature information,so as to better assist the SCH diagnosis.In conclusion,this study based on the information transfer network,SCH abnormal patterns can be mined from the perspective of information transfer mode,information transfer intensity and network topology,indicating that these abnormal patterns are an important cause of SCH episodic memory disorder.Finally,a classification model was constructed to classify SCH.The results show that the information transfer network combined with resting state and task state can provide more effective and comprehensive feature information,so as to better assist the diagnosis of SCH.