Structure-function correspondence in the cortical organization of language

Abstract
Characterizing the organization of the language system in individual brains has long been a challenge in both
research and clinical practice. Advances in fMRI-based localization of functionally specialized brain regions
have revealed that the language areas in individual brains are highly focal and reliably selective in their unique
response to language. However, cortical areas are also defined by their unique structural characteristics, and
these are not well understood for language areas. To address this knowledge gap, the proposed work will use
a structure–function predictive framework to accurately localize language areas in individual brains. This
approach will test the hypothesis that cortical language areas are defined by distinct structural features. Aim 1
and Aim 2 both use local architectural features and long-range connectivity profiles to predict the location of a
fMRI-defined cortical language area. In Aim 1, an interpretable regression model is used to identify the
structural features supporting language computations in the specialized cortical areas. With Aim 2, we instead
use a more complex model to localize language areas with very high accuracy, supporting the existence of
structure–function correspondence while also providing a useful localization tool with far-reaching applications.
Aim 1 and Aim 2 reflect distinct approaches to understanding structure–function correspondence. In Aim 2, the
model is capable of extracting complex feature representations and combining information across levels of
granularity, which impedes interpretation of the important structural features but supports high accuracy
classification. Lastly, in Aim 3, the predictive models are applied to individuals with typical or impaired
language skills to understand how cortical arealization relates to interindividual differences in the capacity for
language. We hypothesize that accuracy of the structure–function predictions will be significantly associated
with language test scores, suggesting a relationship between cortical arealization and behavior. This work will
contribute to research and clinical work by (i) providing structural evidence supporting the existence of cortical
language areas, (ii) developing new methods for localizing the language network, and (iii) advancing our
understanding of the relationship between cortical arealization and language competence.

Grant Number: 1F31DC022801-01
NIH Institute/Center: NIH
Principal Investigator: Rebecca Belisle