Effects of Preprocessing
Steven Strother
University of Minnesota
Radiology
I argue that published findings suggest that new insights into human brain function may be obscured by suboptimal or overly restricted choices in the data-processing pipeline. I review work on performance metrics for optimizing processing pipelines, including prediction, reproducibility, and related empirical Receiver Operating Characteristic (ROC) curve metrics.
Using the NPAIRS split-half resampling framework for estimating prediction and reproducibility metrics (Strother et al., NeuroImage, 2002; LaConte et al., NeuroImage, 2003; Shaw et al., NeuroImage, 2003), I illustrate their application by examining the relative importance of selected pipeline components—interpolation, in-plane spatial smoothing, temporal detrending, and between-subject alignment—in the group analysis of BOLD fMRI scans from 16 subjects performing a block-design parametric static-force task.
Both prediction and reproducibility metrics were necessary for optimizing the pipeline and yielded somewhat different conclusions. In addition, parameter settings for individual pipeline components interact in complex ways. Therefore, the common practice of optimizing components in relative isolation is unlikely to produce fully optimized data-processing pipelines.
