Spatial Craving Patterns in Marijuana Users: Insights from fMRI Brain Connectivity Analysis with High-Order Graph Attention Neural Networks

TL;DR

An elucidative framework termed high-order graph attention neural networks (HOGANN) for the classification of Marijuana addiction, coupled with an analysis of localized brain network communities exhibiting abnormal activities among chronic marijuana users, demonstrates superior performance in cohorts exhibiting prolonged dependence.

Abstract

The excessive consumption of marijuana can induce substantial psychological and social consequences. In this investigation, we propose an elucidative framework termed high-order graph attention neural networks (HOGANN) for the classification of Marijuana addiction, coupled with an analysis of localized brain network communities exhibiting abnormal activities among chronic marijuana users. HOGANN integrates dynamic intrinsic functional brain networks, estimated from functional magnetic resonance imaging (fMRI), using graph attention-based long short-term memory (GAT-LSTM) to capture temporal network dynamics. We employ a high-order attention module for information fusion and message passing among neighboring nodes, enhancing the network community analysis. Our model is validated across two distinct data cohorts, yielding substantially higher classification accuracy than benchmark algorithms. Furthermore, we discern the most pertinent subnetworks and cognitive regions affected by persistent marijuana consumption, indicating adverse effects on functional brain networks, particularly within the dorsal attention and frontoparietal networks. Intriguingly, our model demonstrates superior performance in cohorts exhibiting prolonged dependence, implying that prolonged marijuana usage induces more pronounced alterations in brain networks. The model proficiently identifies craving brain maps, thereby delineating critical brain regions for analysis

Figures (9)

• Figure 1: The overall framework of the HOGANN utilizes two sub-models to perform fusion learning on fMRI time series. The high-order graph attention neighbor mixing model combines hopping to conduct high-order attention for message passing. Sequential graph attention learning uses GAT-LSTM to learn the temporal sequence graph and discern differences between instances.
• Figure 2: The proposed first sub-model contains MixHop and a graph attention layer, wherein the second layer of message passing, concatenating different powers of $\hat{A}^{j}$, can enhance the depth of attention in the neighborhood node connectivity in the pathway.
• Figure 3: The impact of KL divergence ablation studies on mixed embeddings $\textbf{H}^{a}$ and $\textbf{H}^{\prime}$: (Left) Comparing KL divergence with and w/o MixHop. (Right) Comparing KL divergence with weighted loss versus unweighted loss.
• Figure 4: The ablation analysis of HOGANN classification in LM and HC classification. Panel (A) demonstrates the performance changes with multiple attention heads, while Panel (B) illustrates the classification performance changes when the neighborhood K in of the HOGANN is varied. Panel (C) comparing different models based on their classification performance on the marijuana-323 with varying multi-segment time window sizes.
• Figure 5: A comparison of machine learning methods with our HOGANN in identifying the optimal community clustering. In Panel (A) and (B), we compare community detection in weighted functional connectivity matrices using machine learning and our HOGANN, where we map the most significant four communities of the LM group to their corresponding brain regions of the connectivity network. Based on Panel (C) for the LM group, we depict four communities corresponding to the colored boxes, illustrating the connectivity of the most active brain regions in Panel (D).