This preprint examined whether cortical locations identified as critical for speech and language by direct electrocortical stimulation (ECS) exhibit distinct fMRI-derived network signatures. The investigators studied eighteen participants who performed five separate language tasks while undergoing functional MRI. Cortical sites were labeled as critical when ECS produced either speech arrests (SA) or language errors (LE); these critical sites were compared with non-critical sites using graph-theory network measures computed from task-based fMRI data.
The authors computed local and global connectivity metrics as well as measures of cross-community or between-network connectivity. Analyses considered whole-brain connectivity and also evaluated metrics restricted to smaller regions. Across comparisons, both types of ECS-defined critical sites—SA and LE—showed lower local and global connectivity relative to non-critical sites. That is, critical sites tended to be less integrated at both the immediate/local level and across the overall network than sites that did not produce behavioral disruptions when stimulated.
Sites where stimulation induced speech arrests were distinct from LE and non-critical sites in their network profile. SA sites shared the reduced local and global connectivity observed for critical sites overall. However, SA sites did not show the same elevated cross-community connectivity seen in LE sites. In other words, SA sites appeared less connected both within local neighborhoods and to other communities across the brain, lacking the connector-like profile that characterized LE sites.
Cortical sites where ECS produced language errors displayed a pronounced connector profile. LE sites exhibited greater connectivity across sub-networks (communities) than both non-critical sites and SA sites, consistent with a role as bridges linking functional modules. This cross-community connectivity was most apparent when metrics incorporated whole-brain connectivity patterns; the distinctive connector property of LE sites diminished when analyses were limited to connections within smaller regions. The findings suggest that LE sites are functionally positioned to coordinate information across distributed networks involved in language processing.
Connector-type LE sites were not uniformly distributed across cortex. The authors observed concentrations of these cross-network connector sites primarily in temporal regions, including the temporal pole and the superior and middle temporal gyri, as well as in inferior parietal cortices. The spatial pattern aligns with language-related processing regions and supports the interpretation that connector topology in these areas contributes to language function and to the behavioral effects observed during ECS.
The study also leveraged network connectivity features to train machine learning models aimed at predicting which cortical sites would be labeled critical by ECS. According to the abstract, these models accurately predicted critical sites based on the fMRI-derived connectivity measures. The source does not report specific performance metrics, cross-validation details, feature sets, or model types in the abstract; those methodological details were not provided in the portion of the report supplied here.
These findings extend prior electrocorticography (ECoG) results by demonstrating that fMRI-derived whole-brain connectivity patterns relate to the criticality of cortical sites for speech and language. The differential network signatures of SA and LE sites—particularly the connector-like role of LE sites—support a view that behavioral effects from ECS depend on a site’s embedding in distributed functional networks rather than solely on local physiology. The authors propose that network-derived features could provide a framework for predicting cortical regions critical for speech and language, with potential to accelerate or augment invasive mapping procedures used in surgical planning. The abstract notes these conclusions are based on the reported analyses; supplementary methodological and performance details were not reported in the provided source text.
The supplied source text is an abstract and does not include full methodological parameters, sample characteristics beyond the count of eighteen participants, statistical values, or detailed machine learning results. Specifics regarding task paradigms, preprocessing pipelines, graph metric definitions, and model validation procedures were not reported in the abstract and therefore are not described here.
In this cohort, ECS-defined sites critical for speech and language exhibited distinct fMRI-derived network properties. Both SA and LE sites had lower local and global connectivity than non-critical sites, while LE sites uniquely showed increased cross-community connectivity consistent with connector hubs concentrated in temporal and inferior parietal regions. The authors demonstrate that these connectivity signatures can be used to predict critical sites, suggesting whole-brain functional topology contributes to the behavioral consequences of cortical stimulation and may inform clinical brain mapping strategies.