The Study Of SVM-based FMRI Data Classification On MCI

MCI conventional inspection methods have long period and large errors. With the development of functional imaging techniques, fMRI is gradually applied to the diagnosis of MCI, but the diagnosis is still need to rely on the doctor’s experience, it is difficult to promote. Using data mining techniques analyze fMRI data, the diagnosis model can better assist clinical diagnosis. How to extract classification feature of MCI to build classifiers so is the key to building diagnostic model.In this study, based on existing research, studies of MCI feature extraction and classification model. First, extract the BOLD effect as a single voxel classification features, and then use SVM method to constructed single voxel weak classifiers. Filtering high accuracy single voxel classification, build integrated classifier by AdaBoost method, used to assist clinical diagnosis of MCI. The main work is as follows:Firstly, analysis FMRI data features of MCI, extract single voxel’s BOLD curve characteristics.Secondly, use SVM classification algorithm to construct multiple weak classifiers, and by LOPO methods to test the accuracy of classification and analysis of high classification accuracy voxel distribution area. Experimental results show that the voxels correct areas with higher accuracy are key brain areas of MCI.Thirdly, based on voxels with higher classification accuracy, this study used AdaBoost method to establish an integrated classifier. The average accuracy rate reached to80%; On this basis, further analysis of the generalization ability, the experimental results show that this method has good generalization ability. In summary, the results of this study fully proved the feasible method of fMRI data classification through SVM, effective auxiliary MCI diagnosis, has a certain reference and application value for the study of the relevant cognitive disorders.