Recovery-phase contamination of the resting baseline window in block-design fNIRS: a computable criterion and a diagnostic protocol

Abstract

In block-design functional near-infrared spectroscopy (fNIRS), the resting baseline window is placed by convention rather than by derivation. We show that the recovery phase of the preceding hemodynamic response and anticipatory recruitment for the upcoming task act on that window in opposite directions, and that when rest intervals are not symmetric across conditions this asymmetry is sufficient to fabricate a condition difference. Using noise-free simulations built from real block sequences, in which the true coefficient for upcoming load is exactly zero by construction, design structure alone displaced the method zero point in 75 of 80 configurations spanning four regression settings, five amplitude-generating models, two filter states and two samples; the existence of the spurious effect did not depend on the amplitude model, whereas its sign and size did. Recovery times measured under a single criterion in four datasets from three laboratories were 9.2 s, greater than 14.2 s (right-censored), 22.6 s and 16.1 s, and were not proportional to block length, so no universal "rest N seconds" rule is available. A coded sample of 88 block-design fNIRS papers (2021-2026, Europe PMC open-access full text) found that 47.7% did not specify the baseline window reproducibly, that no paper anchored it solely after the preceding block ended, and that none reported a sensitivity analysis for its placement. We therefore provide what a fixed rule cannot: a computable criterion whose terms are measurable in any design, together with a diagnostic protocol and a worked demonstration on an adult ADHD dataset. On a published open dataset (Shin et al., 2018), holding the sample and all other processing fixed and changing only the definition of the baseline window turned an activation estimate indistinguishable from zero (+0.017 microM, p = .55) into a highly significant one (+0.109 microM, p = 1.3 x 10^-6). NOTE ON THIS RECORD: This deposit fixes the state of the manuscript and its supporting materials as of the deposit date, for the purpose of establishing a dated record of authorship prior to circulating the work for advisory comment. The manuscript body is currently in Chinese; an English version is in preparation for submission to Behavior Research Methods. Files are restricted; access will be granted on reasonable request. No participant-level data are included in this deposit.