Correction of connectivity induced by autocorrelation and filtering in resting state functional near-infrared...

Seed-based correlation is commonly used to assess resting state functional connectivity in functional near-infrared spectroscopy (fNIRS). The sample independence assumption underlying correlation is violated in both unfiltered and filtered fNIRS data, due to the non-white frequency spectra of acquired fNIRS signals. This violation inflates the false positive rate, induces spurious correlations, and consequently alters connectivity results.

We extend an existing correction framework originally developed for fMRI, to accommodate the coloured spectral properties of fNIRS signals. We derive the distribution of sample correlation for non-white fNIRS signals and analytically establish that both signal autocorrelation and temporal filtering reduce the effective degrees of freedom. A statistical correction is derived for both Fisher's z-transformation and Student's t-test, incorporating spectral modelling of the ensemble variance across channels.

Simulation results demonstrate that the proposed correction restores valid statistical inference for both unfiltered and filtered coloured signals. Applied to two experimental fNIRS datasets, the correction substantially reduces spurious connectivity, yielding strong and specific bilateral connectivity patterns consistent with known resting state connectivity networks.

Pre-whitening addresses autocorrelation by modifying the signals directly, which may alter connectivity patterns, whereas the proposed correction adjusts the degrees of freedom of the statistical test, leaving the signals intact and requiring only the computation of ensemble spectral variance.

The proposed correction accounts for the non-white frequency spectra of fNIRS signals and the effect of temporal filtering, restoring valid statistical inference in seed-based connectivity analyses. This correction can be incorporated into fNIRS processing pipelines.