Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Stimulus-response correlation space analysis dissociates spatiotemporal cortical networks supporting speech production
Poster Session C, Saturday, September 13, 11:00 am - 12:30 pm, Field House
Arka Mallela1,2, Raouf Belkhir1, Eliza Reedy1, Thandar Aung3,4, Catherine Liegeois-Chauvel1, Hussam Abou-Al-Shaar1, Jiahao Chen1, Julien Dirani1,5, Jorge Gonzalez-Martinez1,4, Bradford Mahon5,6; 1University of Pittsburgh School of Medicine, Department of Neurological Surgery, 2Rush University Medical Center, Department of Neurological Surgery, 3University of Pittsburgh School of Medicine, Department of Neurology, 4University of Pittsburgh Epilepsy Center, 5Department of Psychology, Carnegie Mellon University, 6Neuroscience Institute, Carnegie Mellon University
Introduction: The spatiotemporal distribution of cortical activation during language production is a central question in cognitive neuroscience. High spatial/temporal resolution recording over multiple brain regions can help resolve this question. However, to interpret this electrophysiological data, specific psycholinguistic manipulations with testable behavioral predictions are necessary to separate neural variance attributable to dissociable processing stages. Here, we combine a delayed visual naming paradigm in conjunction with intracranial electrophysiology to identify specific spatiotemporally separated cortical networks supporting specific stages of speech production. Methods: Subjects (20 healthy, 15 intracranial electrophysiology) are presented visual stimuli (pictures, words, pseudowords) to name but are instructed to delay response until a go cue at 0/400/1000ms. Behavioral analyses examine the effect of interaction of psycholinguistic factors (lexical frequency, phonological density, etc.) and delay on response time (RT). For electrophysiological analysis, we exploit the temporal variance induced by delay. We calculate stimulus/response-locked trial-to-trial correlation of high gamma (70-200Hz) activity at each contact/region and project this to a stimulus-response correlation space. We parcellate cortical networks in time and space by shifts in stimulus-response correlation space in response to psycholinguistic factors and compare these changes in RT induced by the same factors to validate separation of networks. Results: Behaviorally, the factor delay modulated the effect of lexical selection on RT in both populations (F6,3176=2.49, p=0.021) but did not modulate temporal parameters of phonological planning (F2,3182.6=0.92, p=0.04), confirming the paradigm separates processing stages. Stimulus-response correlation analyses revealed globally invariant structure across subjects. Certain regions showed strong stimulus-locking (occipital lobe) or response-locking (motor cortex), but most regions were intermediate. Distances in stimulus-response correlation space did not correlate with physical (Euclidean) distance between contacts in the brain (R2=0.013), suggests that our approach was not simply capturing brain structure. Shifts in correlation space identified dissociated cortical networks that predicted substantial variance in RT from lexical selection (R2=0.64) and phonological density (R2=0.75). Using a leave-one-subject-out cross validation approach, these shifts explained 31.4% of variance in RT at the trial level. We then examined the activity of these networks in time for various stimuli categories and examined the shifts in cortical networks in amplitude and peak time. We find that there are sequential but overlapping stages of speech production widely distributed across the cortex. Conclusion: Stimulus-response correlation space reveals stable spatiotemporal signatures of psycholinguistic processing that generalize across individuals. Analysis in stimulus-response correlation space provides a way to leverage temporal variance in the task to separate processing stages temporally and across cortical locations. This approach also provides a functional space in which electrophysiological data from multiple subjects, each with different electrode coverage, can be directly compared. Using this approach, we identify a mixed serial and overlapping process of speech production.
Topic Areas: Language Production, Computational Approaches