Cardiovascular pulse arrival

The time of the cardiovascular pulse arrival in brain was measured from the ECG-verified QRS-complex event timing and the pulse dip in the MREG signal measured at the beginning of A3 segments of anterior cerebral arteries (ACA) defined at two neighbouring voxels centred at midline MNI coordinate (0, 30, −3). An example for the signal shape is shown in Fig. 1A, with additional details in Fig. 2A. Consistently in each cardiac cycle, we observed an MREG signal dip in the ACA following the QRS. The consistency of arrival was visually confirmed with narrow band cardiac MREG data (0.7–1.5 Hz), and the more accurate timings were measured with the wide band cardiac data (0.6–5.0 Hz) used in all further analyses (Fig. 2A).

The cardiovascular pulse propagation vector and voxel-wise pulse arrival times relative to the ACA. (A) Schematic overview of cardiovascular pulse propagation in human brain as functions of time and spatial location. The cardiovascular impulse induces a sharp drop in the MREG signal that moves through the brain as a wave. Optical flow algorithm follows this drop to calculate the local propagation speed (see ‘Materials and methods’ section’. (B) Mean cardiac pulse arrival times with ACA reference [MNI (0, 30, 3)] in seconds for Alzheimer’s disease (AD) and control groups. The bottom row shows their difference (P < 0.05, FDR-corrected), where lower pulse arrival latency is shown in purple (AD early) and greater latency is shown in turquoise (AD late).

Examples of the MREG signal shape. (A) Physiological MREG signals at the ACA in QRS synchronization with simulated ECG plots. From top to bottom: simulated ECG signal matching measured QRS timing, wide band cardiac MREG signal used in this study (0.6–5.0 Hz), narrow band MREG cardiac signal (0.7–1.5 Hz), and respiratory MREG signal (0.15–0.5 Hz). Vertical axes are in arbitrary units, and only MREG signals are comparable. (B) Cardiovascular MREG signal examples of the 0.9 s cardiac cycle used in this work at several representative locations in the brain midline. Average signals are separately shown for control and Alzheimer’s disease group with 95% CI in the background. Vertical axes are MREG signals in arbitrary units.

We used the signal minimum at the ACA as a trigger to segment individual cardiac cycles in the MREG data. This segmentation was used for all further analyses. As explained in the ‘Latency of cardiovascular pulse arrival in Alzheimer’s disease’ section, we selected beats of 0.9 s duration from 17 393 cardiac cycles (n_CON0.9= 1985; n_AD0.9= 2022; Fig. 3B). For visualization (Fig. 2B), we selected five physiological points of interest in the brain midline and present the mean wide band cardiac signal at those points for the contrast of control subjects and the Alzheimer’s disease group. Figure 2C also shows the 95% confidence intervals (CI), and reveals that different locations have distinct patterns of signal variability (e.g. signal shape and variance). Since our findings were focused around the well-defined pulse arrival time (t = 0.0 s), we also show in Supplementary Fig. 3 that directional changes of pulse propagation in the Alzheimer’s disease group are not dependent on the selected cardiac cycle length.

Speed (v_rms) differences in Alzheimer’s disease and control groups, the distribution of cardiac cycle length, and the whole brain average of propagation speed. (A) The 3D time lapse video of the differences (background colour) and significant differences (P < 0.05, FDR-corrected) between Alzheimer’s disease and control groups in speed magnitude (v_rms) of the cardiovascular impulse propagation in a dynamic 3D plot (see Supplementary Video 1 and Supplementary Fig. 4 for full brain coverage). (B) Median 0.9 s cycles were chosen for optical flow analysis based on the similar heart rate distributions from Alzheimer’s disease subjects and controls. (C) Distribution of mean cardiovascular impulse propagation speed (v_rms) across the whole brain data. Background colours separate the three zones; low (0–14 mm/s, P < 3.0 × 10⁻⁴) and high speeds (44–100 mm/s, P < 1.7 × 10⁻¹⁹) predominate in Alzheimer’s disease, while mid-range speeds (15–43 mm/s, P < 5.1 × 10⁻⁸) speeds predominate in controls.

As shown in Fig. 1B, we defined the voxel-wise pulse arrival latency as the time difference between the signal dip (local minima) at the ACA from that in every other voxel in each cardiac period (Fig. 2C). The search for the local minima was done separately for each cardiac period. First, the cardiac period was extended with a copy of itself, upsampled by a factor of 10 with third-order spline interpolation. Values over the manually selected time threshold of 0.6 s (technically falling after the next QRS signal) were designated as preceding the ACA arrival, i.e. a negative time difference. Each voxel was assigned a timing value (in s) for every 0.9 s cardiac cycle in this manner, and differences were found between controls and Alzheimer’s disease subjects using our extended method based on FSL randomize.35 Since this is circular data, depending on the time threshold by which minima are classified into current or next cardiac cycles (e.g. −0.3 s = 0.6 s in this regard), estimates of latencies lying close to the threshold value will be less accurate due to blurring. Therefore, we focused on data after ACA pulse arrival.