EEG-based Abstinent Heroin Addicts Recognition

At present,heroin addiction brings serious negative impacts on society and irreversible damage to public health.The relevant literature suggests that long-term heroin intake can disrupt brain functions and mechanisms even the addicts are treated with medicines.However,the negative effects of heroin addiction on the brain are not completely known and there is not a clear and simple way to identify an abstinent heroin addict(AHA)and the level of an AHA.Therefore,an objective and accurate evaluation method is needed to achieve the detection of heroin addiction.The electroencephalogram(EEG)used in this paper is a non-invasive technique for measuring electrophysiological signals with high time resolution,and it is an important tool for studying the brain activity of individuals who are at rest or doing tasks.The research in this paper intends to distinguish AHAs and HCs and identify the degree of drug addiction in AHAs by analyzing and processing of EEG data of subjects under the risk decision.At the same time,we propose a combination of an optimal feature selection method and classification algorithm to improve the accuracy of identification,which can be used for the monitoring system for AHAs in the future.In order to simplify the experiment and bring conveniences for the devices,we also analyze and process the resting EEG data and find the effective indicators for the difference between the AHAs and HCs.The specific content is as following:(1)In the identification of AHAs and HCs,because of the large number of features extracted from EEG signals the authors intends to remove redundant items and improve classification accuracy.The authors propose a feature selection method based on the mutual information,it is the max-difference mutual information feature selection(MMIFS)which proposes a formula to calculate the threshold value,to get the optimized feature set.To evaluate MMIFS algorithm,it was tested against feature selection method based on the leave one out cross validation(LOOCV)and four classification algorithms which include the Support Vector Machine(SVM)classifier,k-Nearest Neighbor(KNN)classifier,Naive Bayes(NB)classifier and Logistic regression(LR)classifier.Finally,we can find that the classification accuracy by MMIFS is obviously higher than that by LOOCV feature selection method.In terms of the combination of MMIFS and the SVM classifier,the paper verifies that the superiority of the combination method has the highest accuracy which is up to 98.4%.AHAs can be divided into two categories: severe and mild,according to the weekly dose(g)of heroin addicts and the duration of addiction(years).Then the effective feature set was selected through MMIFS,and the different classifiers were combined to classify two addicts with different degrees.It was found that MMIFS+NB could achieve a high accuracy rate of 87.9%.(2)In order to better realize the goal of providing portable equipment and universalization,this paper analyzes and processes the resting state EEG data.Firstly,we select four nonlinear features to calculate EEG signals of each electrode.Next,the MRCS channel selection algorithm was used to sort the weights of 64 electrodes,and then the first 1,2,3...64 electrode channels were classified and effective electrodes were selected according to the highest accuracy.Finally,different brain regions which have active electrodes were identified in the three relative rhythms respectively.In the theta rhythm,effective electrodes are FC3,CP6,TP7,CP3,F2,F1.In the alpha rhythm,effective electrodes are TP10,F8,AF7,P8,F6,F2,F7,AF8.In beta rhythm,the effective electrodes are FT7,FC5,T7 and TP9.Consequently,accuracy rates can be as high as 79.6%,84.3% and 79.6%,respectively.This paper has a highly potential research value.Firstly,we choose the effective features by modifying the feature selection method and combine the classifier to receive an effective identification of AHAs and their heroin addiction level.Moreover,in order to reduce time cost and simplify experimental tasks,we find the effective channels in theta,alpha and beta rhythms respectively which can make a contribution to a real-time distance learning system and provide information that whether the user is an AHA.