Real-world parameter preferences in task-fMRI pre-processing pipelines

Neuroimaging pipelines for the analysis of task functional Magnetic Resonance Imaging (task-fMRI) data are composedof multiple steps covering data preprocessing and statistical analysis. Each pipeline step requires researchers to makemultiple decisions in order to set the available parameters. In their large-scale study, Botvinik-Nezer et al. [1] invitedresearchers to analyze a shared dataset to answer a set of pre-defined research questions. As a result of this experiment,each team constructed a different pipeline and sometimes reported contradictory results. This flexibility in the parameterschosen by researchers is also referred to as the researcher degree of freedom and the induced differences in the resultsas analytical variability.To examine potential sources of analytical variability resulting from the researcher degrees of freedom, Carp [2]identified the parameter values used in the literature, compared the resulting pipelines and demonstrated pipelinevariations. In neuroimaging research, throughout the years there have been multiple endeavors to encourage codesharing alongside research papers [3,4]. Increasing code sharing effectively creates new opportunities to better understandpipelines by directly examining the source code accompanying scientific papers.The aim of this work is to shed light on the sources of analytical variability. We assess the researcher degrees offreedom by collecting and analyzing a codebase of open source pipelines in AFNI, FSL, and SPM.