Poster D88, Wednesday, August 21, 2019, 5:15 – 7:00 pm, Restaurant Hall
Data-driven meta-analysis of the structural-functional parcellation of temporal and temporoparietal cortex
Alex Teghipco1, Gregory Hickok1;1University of California, Irvine
Speech processing occurs along a hierarchy, with each computational step engaging different yet interacting regions of the brain— from acoustic analysis in primary auditory cortex to phonetic processing, word recognition, and semantic analysis in neighboring higher-order areas. What remains uncertain, however, are the precise locations of these functional subregions. For instance, models of auditory processing diverge on whether phonemes are analyzed in mid-posterior STS or mid-anterior STG. Here, we sought to clarify the areas of the brain dedicated to different aspects of speech processing by taking a data-driven meta-analytic approach that compared patterns of activity in temporal and temporoparietal cortex associated with the 3,107 features (i.e., ~cognitive processes) contained in Neurosynth. By gauging the similarity between activity associated with each pair of features, we extracted a set of abstract, lower-dimensional networks that represent the structure of unique brain-behavior relations embedded in this database.We then compared these gross networks to each other by using them to functionally parcellate the temporal and temporoparietal cortex, and evaluated whether any of them delineated components of speech processing. To facilitate interpretation of these networks, and to validate their differences, we independently parcellated our ROI using its structural connectivity. Lower-dimensional functional networks were extracted by clustering features with affinity propagation, which recursively sends messages between pairs of data points to identify exemplars. Notably, this analysis distinguished activity associated with auditory processes from speech perception, sentence-level comprehension, semantics, linguistics, orthography, word class, and word processing. The main pattern of activity for each cluster was extracted with PCA, and similarity between voxels with respect to these core functional networks was used to cluster voxels. Structural connectivity was used to parcellate the same ROI by performing probabilistic tractography for each voxel in 40 subjects from the Human Connectome Project. Tractograms for voxels were averaged across subjects and overlapping tracts that distinguished portions of the ROI were identified with ICA. Functional networks revealed a number of graded interactions, including a gradual shift from speech and auditory processing in Heschl’s gyrus (HG), to comprehension and speech processing along the ventral STG posterior to HG. Comprehension loaded exclusively on areas anterior to this ventral STG cluster, while posterior areas along the STS, STG and MTG loaded strongly on comprehension and moderately on speech. Remarkably, structural connectivity also dissociated these areas: anterior areas associated purely with comprehension were connected by a pathway spanning the STS/MTG; posterior areas loading mostly on comprehension but also speech were connected to the MTG and precentral gyrus; areas loading on speech and comprehension were part of a pathway connecting mid-STS/ventral STG to most of posterodorsal STG; and areas loading on speech and auditory processing connected to the insula and precentral gyrus. These results indicate that speech perception is associated with patterns of activity spanning mid-posterior ventral STG, and mid-posterior STS/MTG. Further, this structural-functional parcellation illustrates that although a large number of subtle differences in the patterns of activity that are reported in the neuroimaging literature are difficult to interpret, they are grounded in tangible differences between structural networks.
Themes: Speech Perception, Methods
Method: Other