Research On Data Analysis Of Temporal Brain Connectivity Data Based On Machine Learning Technology

As one of the most delicate structures of the human body,the brain has complex functional or structural connections among brain regions.In mental diseases,some patients have organ lesions in the body,and some patients have no organ lesions,but the signal connection between regions of the brain has problems.Nowadays,resting-state functional Magnetic Resonance Imaging is widely used as one of the preferred technology in medical examinations,because it can reflect the activity of various brain regions under different tasks and different time.Many researchers construct brain networks from brain connection data and analyze them,hoping to improve the cognition ability of functional or structural connections between brain regions,and obtain better classification performance.With the development of machine learning technology in the medical field,brain data analysis based on medical images has gradually gained the favor of various researchers.Brain network is one of the research contents that has been paid more attention to in this kind of research.In the study of brain network,it mainly uses the medical images information to construct brain connectivity network,and analyze structure and function of brain connectivity,then extract feature and construct classification models to classifiy brain connection data characteristics.However,the classification of brain data using existing techniques has not reached the accuracy that can be directly applied to real life disease diagnosis.Based on the brain network,this study proposes a framework that can be used for the analysis and classification of brain connection data,improving the classification accuracy of brain connection data features.The main research contents are as follows:Firstly,according to pearson correlation coefficient and changes in BOLD signal,constructing functional connectivity network and dynamic connectivity network,and then constructing connected sequence with temporal characteristics.Calculate the pearson correlation coefficient between each region of interest(ROI)using blood oxygen concentration dependent(BOLD)signals to determine the correlation between each ROI,and use it to construct a functional connectivity network.Based on the functional connectivity network,use the inconsistency of BOLD signal changes to construct a dynamic connectivity network to reflect the changes of BOLD signals in different time segments.Connect all dynamically connected networks based on BOLD signal changes to form fully connected networks,and set a distance threshold between.each ROI to reduce network complexity and form a connected sequence with time characteristics.Secondly,mining ordered frequent sequences and sequence association rules in time-connected sequences.The nodes with a higher frequency of occurrences are mined and the threshold value is set to limit the number of occurrences of nodes to select the most frequent nodes.By connecting and filtering the front and rear nodes of the most frequent nodes and setting a threshold for the number of frequent sequence occurrences,the most frequent sequence is generated.The most discriminant frequent sequence is obtained by analyzing the discrete degree of the most frequent sequence.Then,the Apriori algorithm in the frequent sequence mining algorithm is applied to the time-feature connected sequence.and the association rules of nodes in the time feature connected sequence are obtained by setting the minimum support threshold and minimum confidence threshold.At the same time,to ensure data consistency,frequent sequence features and association rule features are integrated into a matrix through feature combination methods,and the dimensionality of the obtained features was reduced to a reasonable dimension by the PCA dimensionality reduction algorithm.And then the dimensionality reduction features are input into five commonly used classifiers for classification respectively.Thirdly,research on classification methods of brain data based on temporal features.In this study,the characteristic matrix obtained by dimensionality reduction is input into five classifiers respectively for accuracy verification.By comparing the classification accuracy,the classification suitable was selected,and the conclusions of different classifiers were verified.Among them,the SVM classifier has the best performance.By comparing the SVM classification results with current advanced classification methods in terms of accuracy,specificity,and sensitivity,this method obtained better results.Sample analysis was conducted on certain regions and times of some subjects to obtain changes in BOLD signals and the distribution of frequent sequence in the brain.Frequent nodes were extracted and feature distributions were observed to explore the possibility of connectivity abnormalities.This study proposed a method of classification and analysis that can be applied to the characteristics of brain connectivity data.It mainly constructed brain network in brain connectivity data and then mined frequent sequences with temporal characteristics and sequence association rules to formed a feature matrix with temporal characteristics.The characteristic matrix was input into the classifier to select the classifier suitable for this method.This study applied temporal feature into the mining of brain connectivity data feature innovatively,and the feature classification accuracy of brain connection data is 97.8% with support vector machine classifier.In terms of accuracy,specificity and sensitivity,compared with other classification frameworks,the classification effect is better than other current classification algorithms.It provided a new idea and method for the study of brain network,and broadened the research content in the field of medicine and artificial intelligence.