Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Cortical Sites Critical for Speech and Language Exhibit Distinct Functional Network Signatures: Insights from fMRI and Electrocortical Stimulation
Poster Session A, Wednesday, September 30, 11:00 am - 1:00 pm, Wangari Maathai
Birsu Baç1, Robert Flint1, Todd B Parrish1, Matthew C Tate1, Nathan E Crone2, Rick Betzel3, Marc W Slutzky1; 1Northwestern University, 2Johns Hopkins University, 3University of Minnesota
Speech and language are complex processes involving activity in multiple brain regions and networks. Direct electrocortical stimulation (ECS) identifies cortical sites "critical" for these functions to guide surgical resection of malignant brain tissue and reduce post-operative functional deficits. Yet the network properties that distinguish cortical sites critical for speech and language from non-critical ones remain poorly understood. Our recent work using electrocorticography (ECoG) suggests that critical sites occupy distinct roles within functional networks; however, ECoG arrays capture only regional cortical coverage as they are limited by clinical monitoring necessity, thus leaving connections to many areas unobserved. Task-based fMRI offers a complementary approach by enabling network analysis across the entire brain during speech and language engagement. We therefore hypothesized that ECS-defined critical sites exhibit distinct fMRI-derived functional network signatures that are not present in non-critical sites. Eighteen participants with epilepsy or brain tumors performed five language tasks during fMRI acquisition: antonym, reading comprehension, rhyming, silent word generation and picture naming. We defined the nodes of our network as 1) 2-mm radius electrode sites within each participant's ECoG grid and 2) parcels from the 1000-node Schaefer cortical atlas, and edges as correlations of fMRI activity between all node pairs. Electrode sites were either critical or non-critical, with critical sites defined as language error (LE) and speech arrest (SA) sites. We compared six well-known node-based network metrics between critical and non-critical sites: local efficiency (Eloc), clustering coefficient (CC), strength, eigenvector centrality (EC), participation coefficient (PC) and gateway coefficient (GC). Next, we examined the metrics’ anatomical distribution across the whole brain, averaged across participants. Finally, using all six metrics as inputs, we trained support vector machines (SVM) and balanced random forests (BRF) to predict criticality. Critical sites exhibited significantly lower local (CC, Eloc) and global connectivity (EC, strength) than non-critical sites. LE sites also had significantly higher inter-subnetwork or community connectivity (PC and GC) than both SA and non-critical sites, suggesting that LE sites act as connectors among functional subnetworks. This signature of LE sites being connectors was most robust when the connectivity matrix was computed across the entire brain; it was not significant when only connections between electrode sites were used. Further, connector sites (high PC) were primarily concentrated in temporal cortices, and somewhat less in inferior parietal cortices. Finally, these graph metrics enabled significantly greater than chance prediction of critical sites. Rather than being local or global hubs, critical sites are instead less connected overall than non-critical sites. In addition, sites critical to higher language function play important roles as connectors across functional communities. The fact that connector-ness was highest in temporal regions pinpoints the temporal lobe as a key region in facilitating connections across speech and language subnetworks. By characterizing network signatures of critical sites and regions across the whole brain, we provide a framework for predicting cortical sites critical for speech and language function and greater understanding of the effects of ECS on the language network.
Topic Areas: Computational Approaches, Language Production