Abstract
Multivoxel pattern analysis (MVPA) examines the difference in multivoxel patterns evoked by different cognitive conditions in a given region of interest (ROI). At the network level, informational connectivity (IC) is a recently proposed method to assess the functional relationship between two ROIs in terms of multivoxel pattern discriminability. Specifically, in this method, a MVPA classifier is trained for each of the two ROIs, and during testing, the distance between the data point from each trial and the classifier hyperplane is obtained in each ROI and correlated across trials between the two ROIs. This multivoxel pattern based correlation is referred to as IC. We applied IC to investigate signaling pathways during attention control and selection in a cued visual spatial attention task. The results show that (1) along the dorsal and ventral streams, similar visual pathways are activated during both cue processing and target processing, (2) IC is positively associated with behavioral performance whereas the conventional functional connectivity is not, and (3) these results are consistent between two fMRI datasets recorded at University of Florida and at the University of California at Davis using the same experimental paradigm. We conclude that (1) it is important to take into account of multivoxel pattern information when assessing the functional relationship between different brain regions during cognitive processing, and (2) IC, which extends MVPA into the network domain, is robust for this purpose and capable of yielding insights not possible with the conventional functional connectivity in which voxel pattern information is lost through averaging.