Modeling the Brain as an Information Source: An Information-Theoretic Framework for Decoding Cognitive States from fMRI

Abstract

This study explores the hypothesis that the anatomical regions of the brain can be modeled as complementary and interconnected information sources. We propose a novel framework for analyzing the dynamics of these interacting information sources during cognitive tasks using information-theoretic measures. Specifically, we introduce dynamic and static entropy models to quantify the information content within individual anatomical regions, both over time and in relation to specific cognitive demands. Furthermore, we develop two network models based on the dynamic and static Kullback-Leibler (KL) divergence to characterize the regional interactions. Testing our models on fMRI data recorded during Complex Problem Solving (CPS) tasks reveals promising results. Entropy values successfully identify activated brain regions, consistent with the existing neuroscience literature. Furthermore, our Kullback-Leibler network models demonstrate high accuracy in distinguishing between the planning and execution phases of CPS, as well as in differentiating between expert and novice problem solvers. These findings suggest that our information-theoretic approach holds promise for identifying active brain regions, characterizing mental states, and elucidating brain networks associated with cognitive tasks.