Analyzing Resting-State fMRI Data in Marijuana Users via High-Order Attention Brain Network

The sustained use of marijuana significantly impacts the lives and health of people. In this study, we propose an interpretable novel framework called the HOGAB (High-Order Attention Graph Attention Neural Networks) model to analyze local abnormal brain activity in chronic marijuana users in two datasets. The HOGAB integrates dynamic intrinsic functional networks with LSTM technology to capture temporal patterns in fMRI time series of marijuana users. Moreover, we use the high-order attention module in neighborhood nodes for information fusion and message passing, enhancing community clustering analysis for long-term marijuana users. Furthermore, we improve the overall learning ability of the model by incorporating attention mechanisms, achieving an AUC of 85.1% and an accuracy of 80.7% in multigraph classification. In addition, we compare linear machine learning methods and evaluate the effectiveness of our proposed HODAB model. Specifically, we identified the most relevant subnetworks and cognitive regions that are negatively influenced by persistent marijuana use, revealing that chronic marijuana use adversely affects cognitive control, particularly within the Dorsal Attention and Frontoparietal networks, which are essential for attentional, cognitive, and higher cognitive functions. The results show that our proposed model is capable of accurately predicting craving maps and identifying brain maps associated with long-term cravings, and also pinpointing active areas that are important for analysis.