Resting State EEG Study Of Heroin Addicts Before And After Withdrawal Combined Deep Learning

A large number of studies have shown that long-term heroin use can cause changes in the structure and function of the brain.Compulsory withdrawal is one of the main ways to treat drug addiction in my country.It is worth exploring whether the changes in brain structure and function after compulsory withdrawal and prolonging the time of withdrawal can improve the effect of drug addiction.At the same time,it is extremely important to explore a fast and objective identification method for drug addiction that is superior to traditional human biological sample detection.Around the above two issues,the main research content and conclusions of this article are as follows:1.Differences of heroin addicts(AHA1),heroin addicts after short-term withdrawal(AHA2)and normal control group(HC)resting state EEG in 5 frequency bands,63 leads and 7 non-linear features are studied.First,extract the non-linear features of the closed-eye resting EEG of the addiction group,the withdrawal group and the control group,and perform Relief feature selection and SVM classification on the non-linear feature set to obtain the optimal feature subset with the highest recognition rate of AHA1 and HC.Then the brain topographic map analysis and data analysis of the optimal feature subset of the three groups of people are performed.The results showed that the singular value decomposition entropy of AHA1 in the Beta frequency band of the frontal,parietal and temporal regions was significantly reduced compared with HC.And there was no significant improvement after five months of forced withdrawal.Compared with HC,the C0 complexity of the Beta frequency band in the top zone of AHA1 was significantly reduced,and after five months of withdrawal,it returned to the level of the normal control group.Compared with HC,the LZ complexity,Shannon entropy,correlation dimension and C0 complexity of the Gamma band in the temporal region of AHA1 are significantly increased,but after short-term forced withdrawal,the four nonlinear characteristic values of AHA2 are compared with those before withdrawal Significantly reduced and returned to the level of the normal control group.Therefore,from the perspective of nonlinear dynamics,heroin addicts have abnormal brain nonlinear activities.After 5 months of short-term withdrawal,the abnormal nonlinear activities of some brain regions have improved,but the addicts are strange Anomalies in the value decomposition entropy cannot be restored to normal through short-term withdrawal.2.Traditional machine learning to extract EEG signal features is time-consuming and laborious,and the recognition rate of artificial neural networks that directly take EEG signals as input is low.In this paper,a seven-layer network model My Net-7 is designed based on convolutional neural networks,which includes six-layer convolutional layer and a fully connected layer.The convolutional layer is formed by stacking convolution kernels with size of 3×3 and 1×1 alternately.The number of convolution kernels presents a decreasing structure and has fewer parameters than an increasing structure.The My Net-7 network can identify heroin addicts and healthy people with an accuracy of 92.77%,which is higher than 80.27% of VGG-16;it only takes 0.15 seconds on average to predict a single sample,which is much lower than the 3.4 seconds of VGG-16.Therefore,My Net-7 has better classification accuracy and single sample prediction time on AHA1 and HC than VGG-16,which can provide a new option for rapid and effective heroin addiction detection.