A brainnetome atlas-based methamphetamine dependence identification using neighborhood component analysis and machine learning on functional MRI data.

Addiction to methamphetamine (MA) is a major public health concern. Developing a predictive model that can classify and characterize the brain-based biomarkers predicting MA addicts may directly lead to improved treatment outcomes. In the current study, we applied the support vector machine (SVM)-based classification method to resting-state functional magnetic resonance imaging (rs-fMRI) data obtained from individuals with methamphetamine use disorder (MUD) and healthy controls (HCs) to identify brain-based features predictive of MUD. Brain connectivity analyses were conducted for 36 individuals with MUD as well as 37 HCs based on the brainnetome atlas, and the neighborhood component analysis was applied for feature selection. Eighteen most relevant features were screened out and fed into the SVM to classify the data. The classifier was able to differentiate individuals with MUD from HCs with a high prediction accuracy, sensitivity, specificity, and AUC of 88.00, 86.84, 89.19, and 0.94, respectively. The top six discriminative features associated with changes in the functional activity of key nodes in the default mode network (DMN), all the remaining discriminative features are related to the thalamic connections within the cortico-striato-thalamo-cortical (CSTC) loop. In addition, the functional connectivity (FC) between the bilateral inferior parietal lobule (IPL) and right cingulate gyrus (CG) was significantly correlated with the duration of methamphetamine use. The results of this study not only indicated that MUD-related FC alterations were predictive of group membership, but also suggested that machine learning techniques could be used for the identification of MUD-related imaging biomarkers.