Research On Time-frequency Function Connection Of FMRI Based On Fast Multi-dimensional EEMD

With the rapid development of brain science,data analysis of functional magnetic resonance imaging has become an important basis for researchers to understand complex brain functions.The signal measured by this technique originates from the dynamic changes of the deoxyhemoglobin content in the blood of the brain.It is reflected in the data as a non-linear non-stationary signal with complex time-frequency characteristics.Therefore,the brain function network method that is currently constructed is still in the process.During the research phase,the diagnosis of the disease was still performed in the clinic using manual intervention.From the existing literature reports,most of the traditional brain function network researches are analyzed from the time dimension,ignoring the frequency specificity of brain signals when human brains interact.In the research of frequency dimension,researchers are mostly based on subjective experience or classical methods,and lack of uniform frequency division standards,resulting in poor reproducibility of related frequency divisions in the field of brain sciences.At the same time,classical algorithms have more assumptions on data.The ability to fully exploit the signal's own characteristics also hinders further understanding of its intrinsic meaning.In view of the above problems,the present study considers the multi-scale characteristics of functional magnetic resonance signals comprehensively,and proposes a data driving method based on the decomposition of the Empirical Multidimensional Ensemble Empirical Mode Decomposition(EEMD)to combine brain signals from multiple dimensions of time and space.Perform adaptive frequency decomposition and build a multi-scale functional brain network.In this paper,the brain wave signals of normal subjects are firstly decomposed by different oscillation rhythms,and the validity and accuracy of the adaptive multi-scale segmentation method are preliminarily discussed from the perspective of the brain function connection.The experimental data was collected from the Magnetic Resonance Center of the Institute of Psychology,Chinese Academy of Sciences,after advanced cortical pretreatment and the latest meningeal version matching.Then the method was applied to the disease data and the differences in the results of various methods were analyzed and compared.That is,the functional brain network based on the undivided frequency nonweighted network,the multi-scale traditional filter method and the multi-scale fast multi-dimensional EEMD self-adaptive method were constructed in the data of mild cognitive impairment subjects and normal subjects respectively.Finally,different network attributes of different methods are extracted to classify feature sets as classification feature sets.From the perspective of pattern classification,the efficiency of constructing multi-scale brain functional connectivity network classification algorithm framework based on fast multidimensional EEMD method is demonstrated.The main tasks of this study are as follows:(1)Comprehensively analyze the characteristics of functional magnetic resonance imaging signals,and reconstruct the signals from multiple dimensions in time and space.First,perform adaptive decomposition under normal subjects,and extract brain functions that have undergone rigorous correction to connect brain regions for discussion and study.From the perspective of sexuality,it verifies the feasibility and effectiveness of the rapid multidimensional EEMD decomposition method,which provides the basis for the comparison of results of various methods and the classification of disease data.(2)In order to test the feasibility of multi-dimensional EEMD method in multiple datasets,this method was applied in the study of mild cognitive impairment,and the results of different methods were compared.First,the unweighted networks of mild cognitive impairment test subjects and normal subjects under different sparsity degrees were constructed as unscaled single-scale brain networks.Their network properties were calculated and statistical analysis was performed to extract the significant differences without frequency division.Brain network properties.(3)In order to compare the differences in the results of multiple frequency division methods,a multi-scale functional brain network of mild cognitive impairment and normal subjects was constructed using the rapid multidimensional EEMD decomposition method and the traditional filter method,and different methods were extracted.There are significant differences in brain network properties.(4)Construct feature sets of the three method differences attributes,use the Support Vector Machine(SVM)classifier to train the data for classification,compare the classification effect of different methods,and prove that the rapid multi-dimensional EEMD method is introduced into the brain from the perspective of model classification.The effectiveness of the functional network.Comparing the difference attributes of the three methods,it was found that the multidimensional EEMD method is more sensitive to capture the baseline changes of the Blood Oxygenation Level Dependent(BOLD)signal,digging deeper into the nuances of the brain network,and having a more accurate approach to disease classification.Excellent classification effect,which also provides a certain methodological basis for the early diagnosis of the disease.