Abstracts

Rationale: Contemporary models of visual object recognition traditionally emphasize bottom-up, hierarchical visual processes. However, sensory input to the visual system is often ambiguous (e.g. occlusion), which suggests that bottom-up processes may not be sufficient alone to achieve fast visual recognition. Recent work suggests that bottom-up visual processes may be facilitated by top-down input from higher-level frontal regions. Current limitations in spatiotemporal resolution of non-invasive neuroimaging has hindered a more comprehensive investigation of this hypothesis, and as result, the frontal regions responsible for top-down facilitation, and the nature of their interactions with higher-level visual regions, remains unknown. Here, we leverage the high spatiotemporal resolution of intracranial EEG (icEEG) recordings to evaluate global cortical network dynamics during visual object recognition. Methods: : icEEG data were recording from 32 subjects implanted with subdural electrodes (SDEs) as they performed a visual naming task of common objects. To overcome issues related to sparse-sampling, individual subject SDE coordinates were transformed to a common brain space, using surface-based normalization to correct for variability in cortical topology across subjects. A mixed-effects multilevel analysis of broadband gamma power changes (BGA, 60 ?" 120 Hz; with respect to pre-stimulus baseline) was used to identify significant regions of activation during noun naming. BGA time-series were characterized the temporal dynamics of these regions with millisecond resolution. Finally, inter-regional network interactions were characterized using the Short-time direct-Directed Transfer Function (SdDTF), which fits linear multivariate autoregressive models to reveal direct influences (i.e. information flow) between localized brain regions. Results: Within 400 ms. following stimulus onset, two distinct patterns of cortical activity emerged in parallel. In visual cortex, rapid increases in BGA were observed, beginning earliest in posterior occipito-temporal regions. Concurrently, large-scale decreases in BGA were observed across frontal cortical regions. Connectivity analyses revealed that frontal deactivations were coupled with significant decreases (q<= 0.05) in bottom-up information flow (from occipto-temporal to frontal regions). In parallel, top-down information flow (from frontal regions to occipito-temporal cortex) demonstrated significant increases (q<=0.05). Critically, information flow was strongest between orbital frontal cortex and both the fusiform and lateral-occipital cortex. Conclusions: During visual naming, parallel and bidirectional network processes emerge across the cortex. These processes involve a decrease in bottom-up input to frontal language regions, while at the same time these frontal regions increase their top-down influences on higher-level visual regions (fusiform and lateral occipital cortex) in a task-dependent fashion. Together, these findings provide novel electrophysiological support for hypotheses of top-down facilitation during visual object recognition. Funding: This work was supported by the National Center for Advancing Translational Sciences of the National Institutes of Health (5TL1TR000369-08 to CMK), National Institute for Deafness and communication disorders (NIDCD ?" R01DC014589 to NT), and partially by the National Science Foundation (NSF- FO 1533664 to NT). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.