Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

Validation of a Region-Specific Approach to CSF Artifact Correction in Subcortical 7T fMRI

Poster Session F - Tuesday, March 10, 2026, 8:00 – 10:00 am PDT, Fairview/Kitsilano Ballroom

Alexandra Fischbach¹ (), Hallee Shearer¹, Daniel Handwerkerd², Laura Lewis³, Karen Quigley¹, Jordan Theriault^1,4, Ajay Satpute¹, Lisa Feldman Barrett^1,4, Stephanie Noble¹; ¹Northeastern University, ²National Institutes of Health, ³Massachusetts Institute of Technology, ⁴Athinoula A. Martinos Center for Biomedical Imaging

Distinguishing neural signals from physiological noise remains a challenge in neuroimaging, particularly in subcortical regions. Traditional CSF correction pools signals into a single global regressor applied uniformly across the brain, overlooking spatial heterogeneity in CSF-noise and potentially leaving residual contamination. To address this limitation, we developed a local CSF correction procedure that models CSF adjacent to each subcortical ROI. Here, we present a comprehensive validation across task and rest data, comparing local CSF correction with whole-brain regression approaches across thresholds and assessing the impact of motion. We demonstrated that CSF time-series exhibit a cortical-to-subcortical gradient and that local CSF regression (60% threshold) yields stronger resting-state functional connectivity (N=83). We further showed that local CSF regression yields stronger emotion task-based activations in subcortex compared with aCompCor (N=43). To assess robustness, we applied a more stringent 80% CSF threshold and regressed six head-motion parameters from the CSF time-series, then re-computed voxel-wise correlations between CSF voxels and the weighted global CSF average to quantify the contribution of motion to the observed gradient. Motion-corrected and uncorrected CSF signals were highly correlated (r̄ = .94), suggesting that the gradient reflects physiological variability rather than motion artifacts. To evaluate spatial stability, we extracted the top 20% of CSF voxels across maps, and calculated a Dice coefficient, revealing strong agreement (Dice = 0.85). Finally, voxel-wise correlations between CSF and motion parameters clustered near zero, indicating negligible CSF–motion coupling. Overall, these results support that local CSF correction captures meaningful physiological variability and enhances subcortical fMRI signal fidelity.

Topic Area: METHODS: Neuroimaging