Research Of Feature Selection Method Base On Epileptic Brain Network

Epilepsy is a chronic disorder of sudden discharge of neurons in the brain that leads to transient brain dysfunction.Due to the initiation site of abnormal discharge and transfer mode of the different,the clinical manifestations of epilepsy is complex,it contains paroxysmal movement,sensory,autonomic,consciousness and mental disorder.For such disease,previous research mostly is committed to finding the disease physiological lesions,or to construct the network,network or network connection display the abnormal,there is little of the disease data and normal data classification research.Based on the above research gaps,this paper mainly focuses on the application of pattern recognition method to distinguish the MRI data of epilepsy and normal people,and find the connection area,a solid foundation for the prediction and diagnosis of epilepsy.The main work is as follows:First of all,Clinical identification of characteristics of individual differences in brain functional connectivity in patients with epilepsy,we use 60 cases of generalized tonic clonic seizures(GTCS)and 63 cases of normal magnetic resonance data,building large scale brain functional network,using the method of support vector machines,the feature weight sorting,find out the consistent connection area,such as the dorsolateral frontal gyrus.Orbital orbital frontal gyrus,inferior frontal gyrus,middle frontal gyrus,and so on,significant differences in these brain regions in accordance with previous studies,the area with the highest scores for the dorsolateral frontal gyrus,may be an important distinguishing feature of the disease;and confirmed the effectiveness of F algorithm on epilepsy score data classification.Select the F fraction of the top 600 features,the best result for classification,of which three index scores are: accuracy is 81.3%,specificity is 73.33% and sensitivity is 88.89%.The features obtained by the algorithm can be used to classify better classification results.Secondly,Clinical identification of individual differences in amplitude of low-frequency fluctuation in patients with epilepsy,we also use the MRI data of 60 cases of GTCS and 63 cases of normal people,the resting state fMRI data were analyzed,by dividing the slow-5 band and slow-4 band data,and the amplitude of low-frequency fluctuation as the classification feature,using Relief algorithm as the classification algorithm of two sets of data for feature selection,the results are as follows: stable fraction top 500 features,the slow-5 band,the classifier obtained the highest accuracy rate of 80.9%(sensitivity is 73.33%,specificity is 88.89%);in the slow-4 band,the highest classification accuracy rate is 79.6%(sensitivity is 72.63%,specificity is 86.94%).The study found that the three indicators of the slow-5 band were slightly higher than the slow-4 frequency band,indicating that the study of the amplitude of low-frequency fluctuation in the special frequency band is of significance for the study of the disease differentiation.