Learning Dynamic Brain Connectome with Graph Transformers for Psychiatric Diagnosis Classification

Abstract

Graph Transformers have recently been successful in various graph representation learning tasks, providing a number of advantages over message-passing Graph Neural Networks. Utilizing Graph Transformers for learning the representation of the brain functional connectivity network is also gaining interest. However, studies to date have underlooked the temporal dynamics of functional connectivity, which can reflect important markers of brain function. Here, we propose a method for learning the representation of dynamic functional connectivity with Graph Transformers. Specifically, we define the connectome embedding, which holds the position, structure, and time information of the functional connectivity graph, and use Transformers to learn its representation across time. We perform extensive experiments with both non-clinical and clinical resting-state fMRI datasets and show that our proposed method outperforms other competitive baselines in classification and regression tasks based on the functional connectivity extracted from the fMRI data.