Fig 2. Different approaches to optimizing preprocessing pipelines.

This toy example depicts N subjects, with 4 preprocessing pipeline steps; each step may be either applied to data (ON) or not applied (OFF). A standard conservative approach (CONS) applies all commonly-used noise correction steps to fMRI data. A fixed optimal pipeline (FIX) applies the single set of pipeline steps that optimizes average prediction and reproducibility (P,R) across subjects. Individual optimization (IND) selects the combination of pipeline choices, specific to each subject and session, which maximizes prediction and reproducibility (P, R). To see the preprocessing choices for optimized pipelines in our current results, refer to S1 and S2 Tables.