Study On Brain Connectivity Markers In Drug Addictio

Drug addiction is a chronic and relapsing brain disorder that can result in alterations to functional or structural connectivity due to prolonged drug exposure.Investigating brain network connectivity markers for drug addiction is crucial for the quantitative assessment of addiction mechanisms and treatment efficacy.Presently,brain functional connectivity is frequently analyzed using the Pearson correlation coefficient matrix,which is susceptible to interference,potentially affecting the accuracy of connectivity analysis.Moreover,a single functional connectivity analysis may introduce bias.To address these challenges,this study analyzes drug addiction data using machine learning and multimodal brain network analysis methods.The specific details are outlined below:(1)A method employing dynamic sliding window technology to extract sub-time series and a convolutional neural network to extract functional connectivity.This method was validated using resting-state f MRI data from 57 drug addicts and 56 healthy controls,with evaluation metrics surpassing those of FCNN.To verify the model’s effectiveness,traditional statistical analysis methods were employed for static and dynamic research analyses.When compared to traditional statistical analysis and deep learning methods,this approach,along with the dynamic analysis of the traditional method,identified more brain regions related to drug addiction.Experiments were also conducted on different modules,with results indicating that all metrics of this method outperformed those of previous FCNN.This study discovered alterations in brain regions such as the cingulate gyrus,olfactory cortex,hippocampus,amygdala,and insula in drug addicts,suggesting that these biomarkers could offer valuable theoretical support for addiction research.(2)Utilizing a combination of resting-state fMRI and DTI multimodal data fusion,this study employed GRETNA software to extract high-order node networks and combined multiple node network properties to construct feature vectors.These vectors were then input into a support vector machine(SVM)for five-fold cross-validation.The results demonstrated that the average accuracy of the binary classification task using the high-order fusion feature vector constructed from multimodal fusion data reached up to 96%,approximately 24% higher than that of the high-order fusion feature vector derived from a single modality and about 8% higher than that of the multimodal single-feature vector.Furthermore,the study identified significant abnormalities in brain regions such as the orbitofrontal cortex,inferior frontal gyrus,precuneus,caudate nucleus,superior temporal gyrus,rectus,cuneus,and putamen in drug addicts,which were consistent with previous research findings.