| Even if it is safe to pass the detoxification period of the body,severe withdrawal reactions still make it highly likely to relapse.The limitations of the current detection methods for drugs is that cannot estimate the brain damage.Therefore,exploring the damage of heroin addiction to the brain structure and function from the perspective of neuroscience will help to contribute an objective and accurate diagnosis of addiction and formulate an effective treatment plan.But most of the previous studies used a single brain functional network or nonlinear feature,without a comprehensive comparative analysis,the experimental content were not sufficient.To counter the above problems,this thesis adopts the high-low risk decision paradigm,collecting two sets of electroencephalogram(EEG)from 32 heroin abstinents(HAs)and 24 normal controls(NCs).Using brain functional networks and nonlinear features,mining effective indicators of the brain damage,the locations of brain regions that have changed significantly.In addition,combining feature fusion and classification algorithms,a model that can be used to identify and monitor HAs is proposed.The main research work includes the following three points:Firstly,based on the full frequency EEG signals of the two groups,six brain functional networks were calculated separately,combining with five binarization methods,exploring the best combination of brain functional networks and binarization methods.The study found that under the condition of high-risk loss,the coherent brain functional network combined with the 10% ratio binarization method,the recognition ability of the participation coefficient is stronger than the corresponding features under the other conditions,and the feature value is significantly lower than NCs.The results indicate that HAs are prone to loss of cognitive flexibility and emotional restraint when they make high-risk choices but suffer losses.Secondly,extract ?,?1,?2 three rhythms from EEG signal,calculate the optimal feature subset under the brain functional networks and the nonlinear features separately,use statistical testing methods to explore the damage of heroin to different brain regions.The study found that under low-risk benefit,the approximate entropy in the ?1rhythm of the AF8 lead in the right frontal region of HAs is significantly lower than that of the NCs,the C0 complexity in the ?2 rhythm of the Fp2 lead is significantly higher than that of the NCs;under the low-risk loss,the approximate entropy in the d rhythm of the TP7 lead in the left temporal region of HAs is significantly higher than that in the NCs;under high-risk benefit,the singular value decomposition entropy in the d rhythm of the TP7 lead of HAs is significantly lower than that of NCs,the approximate entropy is significantly higher than NCs;under high-risk loss,the participation coefficient in the ?1rhythm in the left central region of HAs is significantly lower than that of NCs.The results show that the brain functional network method has good exploration ability on the left central region damage,and the nonlinear feature method has good exploration ability on the right frontal region and left temporal region damage,two methods have complementary advantages in the study of HAs brain injury areas.Thirdly,perform feature fusion on the two optimal feature subsets under the brain functional network and nonlinear feature research methods,apply five classification algorithms(Naive Bayes,Logistic regression,Support Vector Machine,k-nearest Neighbor,Random Forest),looking for the best model to effectively identify HAs.The fusion features are composed of seven nonlinear features and four brain functional network features.The study found that when using the Random Forest classifier,the highest HAs recognition accuracy rate of 94.64% can be obtained.The result shows that the fusion of brain functional network features and nonlinear features can greatly improve the recognition effect of the target population,and help to more accurately distinguish between HAs and NCs. |
PDF Full Text Request