Functional Connectivity Graph Theory Analysis of Spoken Word Processing Efficiency in Prefrontal Cortical Activation

The purpose of this study was to determine if functional near-infrared spectroscopy (fNIRS), and graph theory analysis of functional connectivity measures derived from hemodynamic changes in the dorsolateral prefrontal cortex (DLPFC) can characterize spoken word processing efficiency in neurotypical listeners. Individual differences were assessed by identifying a low performing (Low-P) and a high performing (High-P) individual. Network science and psycholinguistic models of spoken word recognition predict that word frequency and sublexical phonotactic probability of the word form affects the cognitive processing effort. This novel study assesses predictors of processing efficiency directly using functional connectivity measures of global efficiency, local efficiency, modularity, and hubness , derived from listener's frontal lobe hemodynamic response function during a verbal working memory task.

A total of 20 neurologically typical participants (ages 18-21) completed an auditory working memory task where participants were required to hold words differing word frequency and sub lexical phonotactic probability in memory. Changes in oxygenated (HbO) and deoxygenated (HbR) hemoglobin concentration were recorded with a continuous-wave, multi-channel fNIRS system (TechEn, Inc., Milford, MA) using a 20-channel optode montage across the prefrontal cortex. Partial correlation coefficients were calculated between each channel pair to produce 20×20 functional connectivity matrices. Frontal networks were constructed as a graph where nodes in the graph are the light source and edges are connections (e.g., channels) between nodes. Functional connectivity strength and graph theory measures were examined.

LF and HF words were not processed differently at the behavioral or brain level. Task performance regardless of word frequency was related to brain measures, with higher strength of prefrontal FC relating of worse accuracy (d') regardless of task block, and with high modularity correlating with slow response times on the LF task, measured by HbR signal. Higher efficiencies tended to correspond to better accuracy but none of the tests were significant. High-P showed low FC strength and high efficiency relative to others, while Low-P had high modularity and low efficiency, in line with the direction of the brain-behavior correlations. Lastly, we characterized the centrally important regions ("hubs"). These tended to be located in the left inferior area. High-P's hubs overlapped with those showing consistent hubs behavior at the group level, while Low-P's hubs were uncommon.

We identified network properties related to efficient and inefficient language processing in typical participants, which can be used to assess language function of atypical populations in future studies.