Evaluation of Time-Course-Matched PCA Denoising Techniques in Perfusion fMRI

Motivation: Low signal-to-noise ratio and lengthy TRs limit the complexity and activation detection of perfusion fMRI studies. Goal(s): Our goals were to evaluate a novel local and non-local PCA on 1) task activation detection in a fully sampled dataset and 2) a sub-sampled dataset with fewer task blocks. Approach: We modified the local and non-local DM-PCA from earlier work and evaluated their effects using FSL's FEAT analysis tool on 5 subjects who underwent ASL perfusion fMRI. Results: Local and non-local PCA both significantly improved activation detection, but not for subsampled datasets. Impact: Post-processing techniques, such as our PCA denoising, may mitigate the low SNR of ASL. Increased activation detection may allow studies to use more complex fMRI paradigms or reduce scan time for patient populations.