Pooled-group ICA removes decomposition bias between experimental groups

Evidence

Our data analysis is different from other studies that have used resting-state fMRI to study tinnitus. Often the decomposition of the data into independent component is done by performing two (or more) separate ICA analyses on two (or more) groups (see e.g. (Schmidt et al. 2013 )). This lead to a difficult interpretation since it is then unclear whether group differences represent (true) differences between subject groups, or whether the data-decomposition was slightly different (biases) for the two groups. In contrast, we adopted a strategy where the data is pooled over all participants and data-modalities (restingand fixed-state data). In this way, the data-decomposition is not determined in a potentially different way for both groups, leaving out the necessity of post-ICA matching of corresponding component maps. This allows for a better group-estimate of the components. In addition, by using a dual regression technique we can obtain subject-specific maps and perform group-wise analyses on these maps (Filippini et al. 2009 ). — Dijk, Pim van, Physiology, Psychoacoustics and…, 2703

Categorization

• corroborated: 2 independent passages support this claim — Dijk, Pim van, Physiology, Psychoacoustics and…, 2703
• XREF: connects to: Pooling all subjects’ data in one ICA avoids group-decomposition biases in resting-state fMRI. — Dijk, Pim van, Physiology, Psychoacoustics and…, 2749