High Performance Factoring And Clustering Analysis Of EEG Big Data

The scale of Chinese brain disease patients is huge and continues to grow,and the number of epilepsy patients alone is as high as 9 million.There is an urgent need for the diagnosis and treatment of brain dysfunction diseases.The national "13th five year plan" puts forward to focus on the development of brain disease diagnosis and treatment technology.Electroencephalogram(EEG)can record the electrical signal of brain activity directly.Brain science research and medical practice show that EEG technology is helpful to understand and evaluate the state of brain nerve activity.The development of EEG intelligent analysis technology is an important development direction of brain disease research and diagnosis.With the advent of the era of big data of EEG,big data technology of EEG brings opportunities for the actual diagnosis of brain diseases.The intelligent analysis technology relying on EEG big data can establish accurate subdivision models of brain cognition and dysfunction,which helps to form accurate and personalized evaluation and treatment programs and achieve the goals of early understanding,early intervention and early treatment of brain diseases.At present,the intelligent analysis of EEG big data is still in its infancy,and is highlighted by two major technical challenges:(1)Strong subjectivity.The recognition and division of human brain function depend on the subjective experience analysis of experts.The high complexity,strong noise interference and weak characteristics of EEG signal also aggravate the impact of subjective experience analysis on data calibration,which makes the analysis results of models with strong dependence on calibration data set have subjective factors.(2)Poor adaptability.EEG data intensity varies from person to person(individual difference is great),and it is easy to be affected by environmental factors,which leads to serious deviation and error of EEG intelligent analysis model in different subjects and changeable environments,so it is necessary to manually adjust parameters or even reconstruct the model.In view of these two technical challenges,this thesis proposes the exploration and analysis of EEG big data,without subjective prior experience,mining the potential information of EEG big data,exploring the essential characteristics of brain activity,identifying and analyzing the brain function state.The research is mainly from two aspects:(1)Factor analysis.EEG data contains complex brain activity state and has highdimensional data attributes.The extraction of low-dimensional typical features of highdimensional EEG by factoring helps to recognize different brain activity patterns and explain the working mechanism inside the brain.The large-scale and large-scale characteristics of EEG big data have brought the problem of incalculability,inefficiency and inaccuracy for factorized expression.This thesis proposes a massive parallel Bayesian factorization analysis method(MPBF)for EEG big data.Based on the Bayesian decomposition theory and GPU parallel acceleration technology,the fast factorization of high-dimensional EEG is realized.The experimental results show that the method can decompose the high-dimensional EEG data in arbitrarily dimension without prior information,and has a better performance in data scale and expansion ability.The ability is two orders of magnitude higher than that of the conventional method.(2)Adaptive clustering.Clustering analysis can identify and classify brain function state by mining the essential structure of EEG data.Based on the characteristics of clustering unsupervised algorithm,EEG state is divided without subjective prior label.In this thesis,Clustering by Fast Search and Find of Density Peaks algorithm(FSFDPC)is improved based on the method of global adaptive optimization and linear relationship features.Pattern Search and Fuzzy Criterion Adaptive Clustering algorithm(PSFCAC)was proposed to complete the adaptive clustering of complex EEG data.Experiments show that PSFCAC can automatically select appropriate cut-off distance parameters,automatically determine the number of clusters,and the clustering performance is improved compared to FSFDPC.Based on the above two aspects,this thesis constructed an exploration and analysis framework for EEG big data,and conducted factorization clustering exploration and analysis experiments by epileptic EEG and sleep EEG.Epilepsy state division experiments show that the framework can accurately determine the number of epilepsy EEG states,and the state division accuracy rate reaches 95.95%,which can explore the state of brain nerve activity without prior information.The framework can adaptively classify three and four types of sleep staging states in sleep staging experiments,and the results are consistent with sleep staging standards.The analysis of the above experimental results shows that the framework has strong adaptability to EEG state analysis and can be used to explore and analyze the essential cognition of brain dysfunction.The relevant research in this thesis can be used to solve the problems of strong subjectivity and poor adaptability in the intelligent analysis of EEG big data.