Classification Research Based On The Motion Imaging FMRI Data

The analysis and processing of motor image data is crucial for technologies such as brain-computer interfaces.Compared with the commonly used EEG signals,this article uses fMRI data with better spatiotemporal complexity to perform data mining on fMRI data based on motion imaging,aiming to mine brain function information hidden in motion imaging fMRI data.This article collects data by designing motion imaging experiments and proposes conjectures basing on the brain scientific research methods,and combines different machine learning classification algorithms to classify fMRI data to verify whether conjectures are valid.This article mainly carried out the following work:1.A classification mode based on BOLD time series is proposed for the task state with the same activation level.According to the results of two-sample t-test,the VOIs and the time series are extracted to establish the data set.The PSO is introduced to improve the SVM classification performance.The PSO-SVM model and the decision tree and naive Bayes classification are trained.The classification accuracy of PSO-SVM classifier can reach more than 86%,which confirms the feasibility of this classification mode.2.A classification mode based on dynamic functional connection is proposed for task states with different activation levels and close correlation.The VOIs are defined according to the results of the two-sample t-test.VOIs are dynamically functionally connected and binarized.The GA-RS based on genetic algorithm and rough set theory is used to reduce the data set to establish a data set and construct a PSO.-SVM model,while training decision tree and naive Bayes classifier as lateral contrast.The classification accuracy is improved by about 10%by dimension reduction,and the classification accuracy of PSO-SVM classifier can reach more than 87%,which confirms the feasibility of the classification model.This research combines brain science with data mining technology,the results provide a reference for the positioning of motor imaging functional area,and provide analytical ideas and theoretical basis for the application of fMRI data in brain-computer interface and other technologies.