Robust-tedana: An automated denoising pipeline for multi-echo fMRI data

Abstract

Multi-echo functional Magnetic Resonance Imaging (fMRI) data are acquired by recording image volumes at multiple echo times and can be used to improve the separation of neural activity from noise. TE-Dependent ANAlysis (tedana) is an open-source software tailored to denoising of multi-echo fMRI data. The efficacy of denoising can however be inconsistent, often necessitating manual inspection that precludes its application in large-scale studies where processing is ideally fully automated. Here, we introduce Robusttedana, an optimised denoising pipeline that achieves adequate results at both single-subject and group level. Robust-tedana incorporates Marchenko-Pastur Principal Component Analysis (MPPCA) for effective thermal noise reduction, robust independent component analysis for stabilised signal decomposition, and a modified component classification process. We evaluated its performance on Multi-Band Multi-Echo (MBME) language-task fMRI data from the Australian Epilepsy Project (AEP) using objective measures, comparing to conventional fMRI analysis with and without multi-echo-based denoising. Experts manual evaluation was undertaken on a subset of these data to validate the objective measures. The proposed pipeline both mitigates the prevalence of erroneous attenuation of genuine task activation due to instability of single-subject analysis, and increases the magnitude of group-wise effects. Robust-tedana therefore facilitates advanced analysis of MBME fMRI data in an automated pipeline, including for clinical research assessment of individuals.