Graph Neural Network Reveals the Relationship between Tapping Tasks and the Whole Brain Functional Connectivity

Brain functional connectivity is one of the most important research topics because revealing its relationship with behaviour patterns will significantly help the development of brain science. This paper inherits the idea of resting-state fNIRS research based on graph theory, innovatively uses graph neural networks(GNN) to study the brain functional connectivity under different finger-tapping tasks. This paper proposes a novel model: the task fNIRS analysis model based on graph neural network (tFNIRS-GNN). The model includes two core modules, the graph generation from task fNIRS data and the graph classification by GNN. The tFNIRS-GNN obtains a classification accuracy much higher than the theoretical chance level of the brain information ternary classification, revealing the relationship between different finger-tapping tasks and brain functional connectivity. The experimental results also show the effect of different frequency bands on classification accuracy when extracting features from fNIRS data, consistent with the physiological significance of brain signals.