Simulation results, testing for bias in pipeline optimization.

<p>(A) simulated phantom during baseline (left) and activation (right). (B) median False-Positive Fraction (FPF) as a function of Z-score threshold in null data, with error bars showing the range of FPF across activation loci. The expected null distribution (Normal curve of zero mean, unit variance) is plotted as a gray line. (C–D) ROC curves plotting median TPF vs. FPF over the 16 signal loci (error bars show the range of TPF). Results are shown for 3 PDA pipelines, and a General Linear Model (GLM) reference; the line of no signal detection TPF = FPF is also plotted (dashed line). Fixed pipelines have no preprocessing, and the fixed PC basis is the median basis minimizing distance <i>D</i> from (<i>P</i> = 1,<i>R</i> = 1) across all samples (PC dim. = 1 and 3 for CNR = 1.0 and 0.3). The optimized pipeline and PC basis are chosen to minimize <i>D</i>, separately for each sample.