CMStatistics 2018: Start Registration
View Submission - CMStatistics
B0153
Title: A geometric approach towards evaluating fMRI preprocessing pipelines Authors:  Martin Lindquist - Johns Hopkins University (United States) [presenting]
Abstract: The preprocessing pipelines typically used in resting-state fMRI (rs-fMRI) analysis are modular in nature, as they are composed of a number of separately developed components performed in a flexible order. We illustrate the shortcomings of this approach, as we introduce a geometrical framework to illustrate how later preprocessing steps can reintroduce artifacts that had previously been removed from the data in a prior step of the pipeline. These issues can arise in practice when any combination of common preprocessing steps such as nuisance regression, scrubbing, CompCor, and temporal filtering are performed in a modular fashion. We illustrate the problem using a few concrete examples and conclude with a general discussion of how different preprocessing steps interact with one another. These results highlight the fact that special care needs to be taken when performing preprocessing on rs-fMRI data, and the need to critically revisit previous work on rs-fMRI data that may not have adequately controlled for these types of effects.