Tensor Based MP-PCA Denoising for Multi-Echo Functional Magnetic Resonance Imaging Data

Motivation: Multi-echo fMRI acquisition can be used to perform multi-contrast functional mapping with greater statistical power than single echo acquisition. The tensor structure of multi-echo data can also be exploited for denoising using MP-PCA. Goal(s): To compare MP-PCA tensor based denoising against matrix based denoising. Approach: Optimally combined multi-echo magnitude, susceptibility, standard deviation and tSNR maps from multi-echo EPI data denoised using different matrix and tensor implementations were compared visually and quantitatively in terms of whole-brain tSNR. Results: Tensor based MP-PCA greatly outperforms matrix based denoising, improving the tSNR of combined magnitude maps and quantitative susceptibility maps calculated from the same data. Impact: The tensor structure of multi-echo fMRI data can be exploited to perform tensor-based MP-PCA denoising, which greatly outperforms conventional matrix-based MP-PCA denoising, decreasing standard deviation and increasing temporal SNR. Tensor-based MP-PCA denoising should improve multi-echo fMRI and fQSM studies.