Search Papers | Poster Sessions | All Posters
Poster B148 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
VARX model of the Brain: Functional connectivity dominates response to natural stimulus in human intracranial EEG
Maximilian Nentwich1, Behtash Babadi2, Stephan Bickel1,3,4, Lucas C. Parra5 (); 1The Feinstein Institutes for Medical Research, Manhasset, NY, 2University of Maryland, College Park, 3Zucker School of Medicine at Hofstra/Northwell, Hempstead, NY, 4Nathan Kline Institute, Orangeburg, NY, 5The City College of New York
When ongoing sensory stimulation reaches the brain, the activity that it generates reverberates across recurrent brain networks. To distinguish the effect of the stimulus from the recurrent brain dynamic we propose a vector-autoregressive model with external input (VARX). This model decomposes the traditional multivariate "temporal response functions'' into an immediate response to the stimulus, and a recurrent dynamic within the brain. The model naturally combines with the statistical formalism of Granger to determine statistical significance of each of these effects, including direction of effects between brain areas based on temporal precedence. We demonstrate the approach with intracranial EEG (iEEG) recordings of human subjects watching movies (22 patients). We find that the "recurrent connectivity" during the movies is nearly identical when subjects are at rest. The recurrent dynamic enhances and prolongs the responses to sound, scene cuts and other external events. If these external inputs are not accounted for, they induce spurious connectivity between brain areas. The recurrent connectivity differs from conventional “functional connectivity” in that it is directed and asymmetric. We find that sensory areas have mostly outward connections, whereas higher-order areas have mostly incoming connections. The proposed VARX model combines notions of functional connectivity with stimulus response, potentially resolving conceptual problems that arise when discussing temporal integration or prediction whilst ignoring the recurrent dynamic of the brain. Code for the VARX-Granger formalism is available here https://github.com/lcparra/varx.
Keywords: intracranial EEG Granger-causal functional connectivity temporal response function