Characterizing systemic physiological effects on the blood oxygen level dependent signal of resting‐state fMRI in time‐frequency space using wavelets

FIGURE 1. (a) Calculation of heartbeat interval (HBI) and respiration volume per time (RVT) timeseries from raw pulse oximetry or respiratory bellow recordings. (b) Time‐averaged cross‐wavelet power of HBI/RVT and BOLD activations for each resting state network across frequencies.

FIGURE 2. Wavelet transform coherence (WTC) plots from a representative subject. WTC visualization highlights different frequency bands of visual network coherence with heartbeat interval (~0.08 and 0.25 Hz) and respiration volume per time (~0.02 Hz). Color bar signifies magnitude of coherence measured in units of R 2. Arrows indicate phase offset between the two temporal signals. Outlined areas represent time‐frequency points with significant coherence, as determined by a Monte Carlo simulation with boot‐strapped timeseries.

FIGURE 3. Coherence between heartbeat interval (HBI) and resting‐state network (RSN) blood oxygen level dependent (BOLD) signal activations. (a) Percentage of time with significant coherence between RSN BOLD activations and HBI was averaged across participants for each frequency. Coherence between RSN BOLD activations and a null HBI signal, with the same power spectrum as true HBI, are represented by dashed lines. Standard error bars shown in Figure S1a. (b) Percentage of time with significant coherence was averaged across participants within each frequency band of HBI. RSN profiles were more distinct in the lower frequency band than in the higher frequency band (RP‐ANOVA: low frequency p = 6.5e−17, high frequency: p = 6.0e−4; see Table 1 for post‐hoc paired t‐test results between RSNs within each frequency band). All RSNs exhibited greater coherence with HBI in the lower frequency band than in the higher frequency band (paired t‐test: p < .007 for all RSNs).

FIGURE 4. (a) Percentage of time with significant coherence between resting‐state network (RSN) blood oxygen level dependent (BOLD) signal activations and respiration volume per time (RVT) was averaged across participants for each frequency. Coherence between RSN BOLD activations and a null RVT signal, with the same power spectrum as true RVT, are represented by dashed lines. Standard error bars shown in Figure S1b. (b) Percentage of time with significant coherence was averaged across participants within the dominant frequency band of RVT (0.02–0.04 Hz). The ventral attention, somatomotor, and visual network tended to have greater coherence than the default mode, dorsal attention, frontoparietal, and limbic networks (RP‐ANOVA: p = 7.3e−6; see Table 1 for post‐hoc paired t‐test results between RSNs within the RVT frequency band).

FIGURE 5. Percentage of time with significant coherence by phase offset for heartbeat interval (HBI) and resting‐state network (RSN) blood oxygen level dependent (BOLD) signal activations. Phase offsets demonstrate distinct frequency profiles across networks. Shaded areas represent standard error.

FIGURE 6. Percentage of time with significant coherence by phase offset for respiration volume per time (RVT) and resting‐state network (RSN) blood oxygen level dependent (BOLD) signal activations. Phase offsets demonstrate distinct frequency profiles across networks. Shaded areas represent standard error.

FIGURE 7. Percentage of participants with significant temporal variability of coherence between heartbeat interval (HBI) and resting‐state network (RSN) blood oxygen level dependent (BOLD) signal activations. Significance of temporal variability at each frequency was determined by a Monte Carlo approach with bootstrapped timeseries generated with a vector autoregressive model.

FIGURE 8. Percentage of participants with significant temporal variability of coherence between respiration volume per time (RVT) and resting‐state network (RSN) blood oxygen level dependent (BOLD) signal activations. Significance of temporal variability at each frequency was determined by a Monte Carlo approach with bootstrapped timeseries generated with a vector autoregressive model.