Schedule of Events | Symposia

Scaling Laws in Functional Region of Interest (fROI) Analyses

Poster Session D - Monday, March 31, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Ruimin Gao¹ (rgao76@gatech.edu), Anna Ivanova¹; ¹Georgia Institute of Technology

Functional region-of-interest (fROI) analyses of task fMRI offer enhanced sensitivity and spatial resolution over traditional group-based analyses, but their replicability depends on sample size and parameter choices. We introduce funROI, a new Python toolbox for fROI analyses, and investigate the impact of sample size (N=10 to 500) on three key outputs: (1) group probabilistic maps of significant activation during the localizer task, (2) group-level parcels that serve as anatomical boundaries for subsequent fROI definitions in individual subjects, and (3) effect size estimates of fROI responses to task conditions. Using independent samples from four Human Connectome Project tasks—working memory, social cognition, motor tasks, and language—we show that the correlation between the probabilistic maps generated from different samples is highly consistent at larger sample sizes, with most task contrasts reaching a correlation of 0.9 starting at N=250. The parcel overlap between different samples increases from ~0.6 at N=10 to ~0.9 at N=500, primarily driven by large parcels with significant participant overlap. Parameter choices, such as intersubject overlap at the voxel level and smoothing, substantially influence parcel generation, with different domains (e.g., language vs. vision) benefiting from different parameter combinations. For effect size estimation, even small sample sizes suffice. With the effect size of N=500 as the ground truth, over 90% of N=10 and N=20 samples capture the ground truth within their 95% confidence intervals, improving to >95% with N=50. Overall, this work provides both a computational tool and an analytical framework for robust and replicable fROI analyses.

Topic Area: METHODS: Neuroimaging