Fig. 2.

Single-subject seed correlation analysis of functional connectivity using probabilistic thresholds. For this set of analyses, we compared the effective degrees of freedom (df) produced by the wavelet despiking algorithm (see Fig. 1 and Section 2.4 of the Methods), with the nominal degrees of freedom, (assuming df = N) or the number of time points (a commonly used estimate of df). The subject's EPI image was fully pre-processed and “band-pass” filtered using the MODWT, retaining detail in scales 2–4, representing a commonly analyzed frequency range: 0.02 < f < 0.13 Hz. The df estimate at each voxel was produced by adding the df estimates across scales 2–4. (A) The far left column shows the df maps used by both methods. Seed correlation maps were thresholded at a P value equivalent to FDR q < 0.01. Significant correlations are shown in the figure, along with their corresponding weights (r), for the wavelet-based df estimator (upper) and for nominal df = N (lower). (B) The upper panel shows the distribution of Z scores in the three networks shown in panel A, with the wavelet-based df estimator (dark blue) compared to assuming nominal df = N (light blue). The lower panel shows results of a series of permutations tests designed to estimate the observed vs. expected Type I error for the three seed correlation results in panel A. For any given P value, the observed Type I error (y-axis) should not exceed P (x-axis). The wavelet-based df estimator (dark blue) provided good Type I error control, whereas assuming nominal df = N (light blue) did not.