Search Abstracts | Symposia | Slide Sessions | Poster Sessions
A language network in the individualized functional connectomes of over 1,000 human brains doing arbitrary tasks
Poster Session B, Friday, September 12, 4:30 - 6:00 pm, Field House
Cory Shain1, Evelina Fedorenko2; 1Stanford, 2MIT
A century and a half of neuroscience has yielded many divergent theories of the neurobiology of language [1–10]. Two factors that likely contribute to this situation include (a) conceptual disagreement about language and its component processes, and (b) intrinsic inter-individual variability in the topography of language areas. Recent functional magnetic resonance imaging (fMRI) studies of small numbers of intensively scanned individuals have argued that a language-selective brain network emerges from individualized functional connectomics (iFC) in task-free or task-regressed activation timecourses [11–13]. We tested this hypothesis at scale and evaluated its practical utility for task-agnostic language localization: we applied iFC to the entire history of scanning in the Fedorenko lab: 1,957 (fMRI) scanning sessions (1,199 unique brains), each consisting of diverse tasks. We developed a data-driven network parcellation procedure for arbitrary fMRI data. In brief, within each scanning session, irrespective of task, we used activity correlations between brain regions to compute probabilistic assignments of voxels to putative networks. We then assigned interpretable labels to the networks based on topographic similarity to reference network atlases defined by prior work, including a language network [12,14]. Finally, we studied the functional properties of these networks by comparing their topography to diverse task contrasts within the same individuals across domains, including language, theory of mind, executive functions, and high-level visual processing. To ensure data independence, a critical language localizer task contrasting sentences with nonword lists (S-N) [15] was held out from parcellation. We found that our iFC-derived language network (LangFC) replicated prior task-based claims about the existence and functional properties of an integrated language network [9]. LangFC was similar in topography to the reference atlas for the language network (spatial Fisher correlation or z(r)=0.55±0.002SEM) and highly selective for language, with strong responses to linguistic tasks (S-N: t=2.42±0.03SEM) and weak/negative responses to nonlinguistic tasks (e.g., hard vs. easy spatial working memory: t=-0.51±0.02SEM). LangFC dissociated spatially and functionally from nearby networks (e.g., frontoparietal network A, default network B, and the auditory network) and was more stable within participants across sessions (z(r)=0.80±0.01SEM), than between participants (z(r)=0.41±0.0001SEM), indicating reliable inter-individual variability in structure-function tethering for language. Further analyses showed that LangFC was robust to task state, identifiable from only a few minutes of fMRI data, and successful as a drop-in replacement for task-based localization in reanalyses of a prior experiment [16]. We conclude that iFC reveals a left-lateralized frontotemporal network that is more stable within individuals than between them, robust to task state, and selective for language. These results support the hypothesis that this network is a key structure in the functional organization of the adult brain and show that it can be recovered retrospectively from arbitrary imaging data even without localizer tasks, with implications for neuroscience, neurosurgery, and neural engineering. 1. Geschwind (1970) 2. Hickok... (2007) 3. Hagoort (2005) 4. Friederici (2017) 5. Bornkessel-Schlesewsky... (2013) 6. Matchin... (2020) 7. Pallier... (2011) 8. Brennan (2012) 9. Fedorenko... (2024) 10. Aliko... (2023) 11. Braga... (2020) 12. Du... (2024) 13. Salvo... (2024) 14. Lipkin... (2022) 15. Fedorenko... (2010) 16. Shain... (2024)
Topic Areas: Computational Approaches, Methods