Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Dissociate language selective region from domain general region using functional near-infrared spectroscopy: an individual functional channel of interest analysis approach
Poster A40 in Poster Session A - Sandbox Series, Thursday, October 24, 10:00 - 11:30 am, Great Hall 4
This poster is part of the Sandbox Series.
Haolun LUO1, Qun LI2, Li SHENG1; 1Hong Kong Polytechnic University, 2West China Hospital of Sichuan University
Many cognitive operations in language comprehension are domain-general, suggesting that the "multiple demand" (MD) brain network, associated with domain-general executive functions, may perform these operations (Fedorenko & Shain, 2021). However, recent evidence suggested that the language and MD network are dissociable (Fedorenko et al., 2024). Language-selective regions likely implement these computations locally. Defining functional region of interest is a valid method to dissociate language processing from other cognition in fMRI studies. In this study, we extend this method to fNIRS. Sixteen university students participated in three tasks: an auditory language localizer task (adapted from Scott et al. 2017), a spatial working memory task (adapted from Fedorenko et al. 2011) and listening to a 5-minute narrative recording. In the localizer task, participants listened to intact and degraded audio clips. Each run consisted of two 18s blocks of the intact and degraded audio type, with a 10s fixation at the beginning of each block. In the spatial working memory task, participants tracked four (easy) or eight (hard) locations on a 3 × 4 grid and then chose the correct grid in a two-choice question. The fNIRS measurements were recorded with a 79-channel system. Data were then preprocessed using the HomER3 (Huppert et al., 2009). We first took the standard fixed array analysis for different condition contrasts with Bonferroni correction. Then, in fCOI analysis, we defined the fCOI within participants using a leave-one-run-out procedure (Liu et al., 2022). Specifically, for each participant, one run of data from the localizer task is left out iteratively while the remaining runs are used to define the fCOI. The fCOI identified was used to extract a response for each condition from the left-out run. For each participant, the extracted response were averaged for each condition. We used a linear mixed-effects (LME) analysis on these averaged values to account for individual differences in global signal strength: full model = channel HbO data ~ condition + (1 + 1|subject). We then extracted the HbO time courses when listening to the narrative from the language and MD fCOI. For each participant and fCOI, we computed the intersubject correlation (ISC, Blank & Fedorenko., 2024). We tested whether the average ISCs differed between the language and MD fCOI using LME model: full model = ISCs ~ fCOI + (1 + 1 | subject). For the selected language fCOIs, the mean HbO2 concentration in the left-out data was significantly higher for the intact audio trials than that for the degraded audio (β = -5.20e-7, SE = 2.32e-7, p=0.032). Also, the selected language fCOIs tracked the linguistic input more closely than the MD (β = -0.28, SE = 0.13, p=0.027). However, language fCOI did not show a significant difference between the hard and the easy conditions in the spatial working memory task (β = 9.71e-8, SE = 1.35e-8, p=0.47). In the fixed-array analysis, none of the channels exhibited significantly higher HbO2 concentration in language localizer task. This study establishes the feasibility of identifying language-responsive channels at the individual level by fNIRS with better sensitivity.
Topic Areas: Methods,