Improving the robustness of independent component analysis for denoising multi-echo fMRI data

Multi-Echo fMRI data acquisition has multiple advantages over single-echo acquisition. Principal amongst them is that multiple echoes can distinguish neural activity from artefacts. TE Dependent ANAlysis (TEDANA) is an existing software tool designed to denoise multi-echo fMRI datasets. We evaluated the performance of TEDANA to denoise fMRI data of 120 subjects. Our results demonstrated that TEDANA improved the activation detection at a group level. However, for a subset of subjects TEDANA degraded their individual result substantially. We identified potential causes and proposed a modified framework for multi-echo data analysis that provides reasonable results at an individual subject level.