Integration and segregation in the brain as a cognitive flexibility during tasks and rest

To flexibly respond to a continuously changing environment, the human brain must be able to flexibly switch amongst many demanding cognitive tasks. The flexibility inside the brain is enabled by integrating and segregating information in large-scale functional networks over time. In this study, we used graph theory metrics prior to clustering to identify two brain states, segregated and integrated, in 100 healthy adults selected from the Human Connectome Project (HCP) dataset at rest and during six cognitive tasks. Furthermore, we explored two-dimensional (2D) latent space revealed by a deep autoencoder. In the latent space, the integrated state occupied less space compared with the segregated state. After binning the latent space, we obtained entropy from the probability for each data point of being in the bin. The integrated state showed lower entropy than the segregated state, and the rest modality showed higher entropy in both states compared with tasks. We also found that modularity and global efficiency are good measures for distinguishing between tasks and rest in both states. Overall, the study shows that integration and segregation are present in rest and in task modalities, while integration serves as information compression and segregation as information specialisation. These characteristics ensure the necessary cognitive flexibility to learn new tasks with deep proficiency.