Olivia S. Cheung^1,2 (), Haiyang Jin¹, Lúcia Garrido³; ¹New York University Abu Dhabi, ²Center for Brain and Health, NYUAD Research Insitute, ³City, University of London

Regions of interest (ROI)-based approach is a popular method to study brain activity in cognitive neuroscience research. To what extent are functionally-defined ROIs reliable within individual participants? In a comprehensive, pre-registered fMRI study, we investigated reliability of ROI location and activation profiles across category-selective regions, including bilateral face-selective, body-selective, scene-selective, and object selective regions (FFA1, FFA2, OFA, ATL, pSTS, aSTS, FBA, EBA, PPA, OPA, RSC, and LOC). For each ROI, we examined the reliability of two localization measures: 1) coordinates of peak activation and 2) degrees of cluster overlap, and three activation measures: 1) response amplitudes, 2) selectivity amplitudes, and 3) response patterns. We compared the reliability measures of the ROIs defined within each participant based on either a p-value threshold (p<0.001) or an area threshold of contiguous vertices with the strongest selectivity (≤100 mm2). Because category-selective regions are commonly identified with either images or videos, where videos produce particularly robust activations compared with images in several ROIs, we also compared the reliability measures evoked by either format. All participants (N=30) completed two fMRI sessions viewing images or videos of faces, bodies, scenes, objects, and scrambled objects. Independent data sets were used to define the ROIs and to calculate reliability. Unsurprisingly, we found higher ROI reliability for all measures within- than between-sessions, and reliability was generally higher in the posterior than anterior regions. Notably, reliability was higher for all ROIs defined with images than with videos, with the exception of pSTS and aSTS. Moreover, reliability was consistently higher for ROIs defined with area thresholding than p-value thresholding methods. Together, these findings suggest that the reliability of the category-selective regions depends on specific cortical locations, ROI selection methods, and stimulus formats. Specifically, the use of image presentation and area thresholding to define the ROIs is recommended for most category-selective regions.

Acknowledgements: This work was funded by a New York University Abu Dhabi faculty grant (AD174) and a Tamkeen New York University Abu Dhabi Research Institute grant (CG012).