The study of functional brain networks that support cognitive processes is one of the central goals of modern neuroscience. Functional magnetic resonance imaging (fMRI) is widely used to obtain data on brain activity. However, the high dimensionality and dynamic nature of fMRI data makes their processing challenging. Network-based methods of data representation offer a promising approach to describe the brain as a network, where nodes correspond to brain regions and edges correspond to functional connections between them. This
allows us to further explore the topology of brain networks and their role in cognitive states. The purpose of this paper is to compare ensemble and correlation graphs in a brain state classification task based on functional magnetic resonance imaging (fMRI) data.

Methods. This paper presents a novel method for representing fMRI data in graph form based on ensemble learning. To demonstrate the effectiveness of the data representation method, we compared it with correlated graphs by applying a graph neural network to classify brain states.

Results and Conclusion. Our results showed that ensemble graphs lead to significantly more accurate and stable classification. The better classification performance suggests that using this method we are more efficient in identifying functional connections between brain regions during cognitive tasks.