Model connectivity: overcoming the limitations of functional connectivity by leveraging the power of the voxelwise encoding model framework
Emily Meschke, Matteo Visconti di Oleggio Castello, Jack Gallant, University of California, Berkeley, United States
Session:
Contributed Talks 3 Lecture
Location:
Grand Ballroom A-C
Presentation Time:
Sun, 28 Aug, 10:20 - 10:40 Pacific Time (UTC -8)
Abstract:
Functional connectivity (FC) is a popular method for recovering functional networks from fMRI data. However, because FC does not distinguish between signal and noise in the data, FC networks are biased by noise. Furthermore, networks recovered by resting state FC have no clear functional assignment. To overcome these shortcomings, we developed model connectivity (MC). MC builds on the voxelwise modeling framework (VM), which models fMRI signals in terms of stimulus- and task-related features. Because VM models repeatable aspects of the fMRI data, it separates signal from noise. Because VM is based on stimulus- and task-related features, it provides clear functional assignment. To demonstrate the advantages of MC, we applied it to a naturalistic fMRI dataset in which participants listened to narrative stories (Huth et al., 2016). VM was used to predict brain activity from the semantic content of the stories. MC was used to recover functional networks from the Euclidean distance between the estimated model weights. We find that MC recovers networks that are not biased by noise, and that have a clear functional description.
