MP-PCA denoising of fMRI time-series data can lead to artificial activation “spreading”

• DOI: 10.1016/j.neuroimage.2023.120118
• Publication Date: 2023-04-14T23:49:23Z
• Authors: Francisca F. Fern...; Jonas L. Olesen; Sune N. Jespersen; Noam Shemesh
• Keywords: Human Connectome Project

MP-PCA denoising has become the method of choice for MRI data since it provides an objective threshold to separate signal components from unwanted thermal noise components. In rodents, in coils is important source that can reduce accuracy activation mapping fMRI. Further confounding this problem, vendor often contains zero-filling and other post-processing steps may violate assumptions. Here, we develop approach denoise assess "spreading" caused by rodent task-based fMRI data. Data was obtained N = 3 mice using conventional multislice ultrafast acquisitions (1 s 50 ms temporal resolution, respectively), a visual stimulation paradigm. produced SNR gains 64% 39%, Fourier Spectral Amplitude (FSA) increases BOLD maps 9% 7% data, respectively, when small [2 2] window. Larger windows provided higher FSA with increased spatial extent or not represent real activation. Simulations showed incur false positive rate smoother functional due local "bleeding" principal components, optimal window improved specificity mapping, based on Dice score calculations, depends data's tSNR CNR. This effect applies also another recently proposed low-rank (NORDIC), although lesser degree. Our results bode well enhancing and/or resolution future work, while taking into account sensitivity/specificity trade-offs methods.