Dual regression physiological modeling of resting-state EPI power spectra: Effects of healthy aging

Abstract

Aging and disease-related changes in the arteriovasculature have been linked to elevated levels of cardiac cycle-induced pulsatility in the cerebral microcirculation. Functional magnetic resonance imaging (fMRI), acquired fast enough to unalias the cardiac frequency contributions, can be used to study these physiological signals in the brain. Here, we propose an iterative dual regression analysis in the frequency domain to model single voxel power spectra of echo planar imaging (EPI) data using external recordings of the cardiac and respiratory cycles as input. We further show that a data-driven variant, without external physiological traces, produces comparable results. We use this framework to map and quantify cardiac and respiratory contributions in healthy aging. We found a significant increase in the spatial extent of cardiac modulated white matter voxels with age, whereas the overall strength of cardiac-related EPI power did not show an age effect.

Keywords: Cardiac pulsatility; Dual regression; EPI; Healthy aging; Physiological noise; fMRI.

Fig. 1

Exemplar whole brain rs-EPI power spectrum (red) and the power spectra of the cardiac (blue) and respiratory trace (green) from external measurements (pulse-ox at the finger and a pressure pad on the chest). The external physiological signals are sub-sampled to the EPI TR. Above the low-frequency neurovascular range ($\mathrm{\lesssim }0.1$ Hz) the EPI spectrum closely matches the physiological spectra. The cardiac spectrum shows aliases of the second and third harmonics of the dominant cardiac frequencies (according to $f_{\text{alias}}=|f_{\text{true}}-n\cdot f_{\text{sample}}|,n\in \mathrm{\mathbb{N}}$).

Fig. 2

PE maps from an exemplar subject. The noise baseline map α in (a) fills the entire brain as every voxel has a constant thermal noise level. The respiratory map (b) is more pronounced in the periphery and the cardiac map (c) is more prevalent around arterial structures and narrow CSF-filled perivascular spaces. (d) displays the average residual error and (e) is a high resolution anatomical reference T1w image (manually chosen to match the PE maps that are in lower resolution EPI space).

Fig. 3

Iterative dual regression scheme to find brain-specific refined spectra and spatial maps of cardio-respiratory fluctuations.

Fig. 4

Box plots of the percentage power deviation at the dominant frequencies. These are the differences between a) the informed dual-regressed white matter cardiac spectrum and the external pulse-ox spectrum; b) the informed and data-driven dual regressed cardiac spectra in white matter; c) the informed dual-regressed cardiac spectra in grey and white matter; d) the informed dual-regressed white matter respiratory spectrum and the external respiratory spectrum; e) the informed and data-driven dual regressed respiratory spectra in white matter; f) the informed dual-regressed respiratory spectra in grey and white matter. Box plot properties: the red line is the median, the blue box marks the $25\text{\%}-75\text{\%}$ quantiles, the whiskers span the $\pm 2.7\sigma$ range and red crosses mark outliers. P-values refer to Student's paired t-test.

Fig. 5

$R_c$, the ratio of significantly cardiac modulated white matter volume, against age. Results from the informed dual regression are in a) and from the data-driven approach in b). The scatter plots are almost identical and both regressions yield comparable results The p-values refer to a partial correlation with age, controlled for the cardiac rate and the displacement. The r-value is Pearson's correlation coefficient.

Fig. 6

Average distribution of cardiac PE $\langle {\beta }_c\rangle$ in the younger ($19-29$ ys, red plot) and older ($46-77$ ys, green plot) subjects in white matter. The younger subjects' distribution is almost a complete subset of the older subject's distribution (the overlap is coloured in brown). The standard error of the mean is plotted in shaded colour on top of the distributions.

Fig. 7

White matter cardiac β_c maps in MNI space. The left column shows the MNI space slices, the middle column are the group-averaged cardiac maps for the older group and the right column are the younger group. Similar to the distribution of cardiac PEs in Fig. 6, this MNI overlay visualises how the older group has an increased spatial extent of cardiac pulsatility in the white matter.