An Information-Theoretic Analysis of Resting-State versus Task fMRI

Similarities in brain functional connectivity under various imaging conditions have led to resting-state fMRI becoming the preferred alternative to task-based fMRI in many experimental and clinical contexts. However, the equal or superior quality of the data obtained from resting-state fMRI for inferring neural responses is yet to be demonstrated. Here, a systematic comparison of the quality of inferences derived from both imaging paradigms was conducted by means of Bayesian Data Comparison. In this framework, data quality is formally quantified in information theoretic terms as the precision and amount of information provided by the data on the parameters of interest. Parameters of effective connectivity models, estimated from the cross-spectral densities of resting-state- and task time series by means of Dynamic Causal Modelling, were subjected to the analysis. Data from 50 individuals undergoing a Theory-of-Mind task were compared to data from the same individuals undergoing resting-state, both datasets provided by the Human Connectome Project. The examined network was specified from an independent component extracted from the resting-state data and the specific component was selected based on its overlap with the Theory-of-Mind activation map. A threshold of very strong evidence was reached in favour of the Theory-of-Mind condition regarding information gain over connectivity parameters. No notable differences were observed in parameter precision. The active task elicited stronger effective connectivity in this specific network and should be preferred over resting-state in settings where the detectability of the network is crucial, such as in clinical applications.