Study On Word Cognitive States Recognition From FMRI Data

The human brain is the central pivot of the body to gain,store,process,and integrate information both in vivo and in vitro,and to unveil the brain has always been pursued. Millions of bytes of data were resulted by lots of experiments,as new methods and technology were applied to brain research.Bulk and useful information beneath these mass data cannot be excavated only by simple storage,query and calculation.As a result,the methods of data mining and advanced informatics are necessarily performed to analyze, process,integrate,and model the experimental data,to discover new regularity and disclosure further brain cognition.Functional Magnetic Resonance Imaging(fMRI) is an important technology on the cognition research in recent years.The main idea of this paper is to concentrate on the study of data analyzing and processing of fMRI,and to discuss cognitive states recognition methods from fMRI data.The sketch of this work is from three aspects as follows.For the beginning,the research on fMRI image registration is carried out.There has been a problem of fMRI image from a time set experiment cannot be aligned due to subjective and objective reasons.The statistic and analysis are focused on the same area,which made unaligned images lead to bad and useless results.A method of fMRI image alignment based on a new Particle Swarm Optimization(mPSO),which is modified by cloud model,was put forward in this paper,and simulative alignment proved that fast convergence rate and precise aligned images were gained through this method than standard Particle Swarm Optimization.Secondly,feature extraction approach of fMRI images is investigated.Hundreds of thousands of voxels distributed in one image of fMRI experiment,this order of magnitude is the same with the vector form,which severely hindered the subsequent research.For this reason,minimized parameters ought to be utilized to character the primitive image.A fMRI image feature extraction method based on Principal Component Analysis(PCA) developed and minute quantity of Principal Component(PC)(usually below ten) substituted former voxels,what's more,the order of magnitude can be easily controlled by selected PC.Thirdly,a cerebral posi-negative word cognition model is built based on SVM.The experiments showed that 89%average recognition rate was reached when fMRI analysis was applied to single sample,while 83.5%average identification was kept with multiple samples. Compared with the methods shown in published papers,a distinguished advantage in this paper is that the feature vector extraction is independent with the regions of interested,and equal rate of recognition was received without prior information.