Assessing cerebral microvascular volumetric with high-resolution 4D cerebral blood volume MRI at 7 T

Abstract

Arterial pulsation is crucial for promoting neurofluid circulation. Most previous studies quantified pulsatility via blood velocity-based indices in large arteries. Here we propose an innovative method to quantify the microvascular volumetric pulsatility index (mvPI) across cortical layers and white matter (WM) using high-resolution four-dimensional (4D) vascular space occupancy (VASO) and arterial spin labeling (ASL) magnetic resonance imaging (MRI) at 7 T with simultaneous pulse recording. We assessed aging-related changes in mvPI in 11 young (28.4 ± 5.8 years) and 12 older (60.2 ± 6.8 years) participants and compared mvPI with large artery pulsatility assessed by 4D-flow MRI. mvPI peaked in the pial surface (0.18 ± 0.04). Deep WM mvPI was significantly higher in older participants (P = 0.006) than young ones. Deep WM mvPI correlated with large artery velocity PI (r = 0.56, P = 0.0099). We performed test–retest scans, non-parametric reliability test and simulations to demonstrate the reproducibility and accuracy of the method. In conclusion, our non-invasive method enables in vivo fine-grained measurement of mvPI, with implications for glymphatic function, aging and neurodegenerative diseases.

Main

Cerebral arterial pulsation is the rhythmic dilation and constriction of brain arteries with the cardiac cycle, driven by the pressure wave from heart pumping. It was found that higher arterial pulsatility is generally associated with more severe small vessel disease (SVD)^1,2,3,4 and Alzheimer’s disease as well as worse cognitive performance^5,6,7,8, highlighting the role of cerebroarterial pulsatility in the pathogenesis and progression of cerebrovascular and neurodegenerative diseases.

Previous studies attempted to understand the function and underlying mechanisms of cerebroarterial pulsation within the brain. Iliff et al.⁹ proposed that arterial pulsatility is a key driver of cerebrospinal fluid (CSF) movement in the perivascular spaces (PVSs) into and through the brain parenchyma in mice, facilitating the clearance of interstitial solutes and wastes in the brain^10,11,12. In general, arterial pulsation plays a crucial role in driving the glymphatic system^{13,14,15,16,17}, a recently discovered waste clearance system in the brain. Mestre et al.¹⁸ directly observed that hypertension leads to an increase in arterial volumetric pulsatility, which subsequently results in a reduction in CSF flow. This finding underscores the potential influence of alterations in arterial volumetric pulsatility on the glymphatic system and provides a theoretical basis for understanding various cognitive disorders. Additionally, Jammal Salameh et al.¹⁹ found that arterial pulsation can modulate neuronal activity directly via mechanosensitive ion channels. These studies collectively support the role of arterial pulsatility in promoting fluid circulation in the glymphatic system and even directly influencing neuronal activity. However, all these studies were performed in animal models using invasive optical imaging and electrophysiological recording.

In humans, arterial pulsatility is typically measured using transcranial Doppler ultrasonography²⁰ or phase-contrast MRI^21,22 (PC-MRI), mostly in large vessels. Recently, high-resolution (submillilmeter) PC-MRI at 7 T has been applied for measuring the pulsatility of smaller lenticulostriate arteries²³; however, it remains difficult to assess the pulsatility of cerebral microvasculature, including arterioles, capillaries and venules. In addition, existing methods largely quantify velocity rather than volumetric pulsatility, which more directly influences CSF movement.

Here we present a non-invasive approach to assess volumetric pulsatility of cerebral microvasculature across cortical layers in gray matter (GM) and WM, using high-resolution 4D VASO MRI²⁴ with simultaneous recording of pulse signals at 7 T. Changes in mvPI and cerebral blood volume (CBV) across cardiac cycles are assessed through retrospective sorting of VASO signals into cardiac phases and estimating mean CBV in resting state (CBV₀) by ASL MRI²⁵ at 7 T (refer to Fig. 1 and Methods for details). We applied this method to 11 young and 12 older participants and compared mvPI results and pulsatility of large arteries assessed by 4D-flow PC-MRI. We found that mvPI was highest at the pial surface, decreasing toward deeper cortex and larger arteries. In deep WM, mvPI was substantially higher in older adults. Test–retest, non-parametric analyses and simulations confirmed the reproducibility and accuracy of our approach.

Fig. 1: Schematic diagram of mvPI calculation process.

a, Cardiac phase definition: two complete cardiac cycles of pulse signals are shown. The time from one cycle’s systolic peak to the next systolic peak is divided into 10 equal phases, defined as phases 1–10. b, Retrospective sorting of images into 10 cardiac phases. The time and signal intensity curves of VASO (red) and BOLD (blue) signals are shown. The lower figure shows the trigger time corresponding to each image and its corresponding cardiac phase (black). c, CBF and CBV₀ laminar profiles based on high-resolution (iso-1.25 mm) ASL scans and CBV₀ were derived based on the listed equation. d, Variations of VASO signals across cardiac phases after retrospective sorting of VASO images; the mvPI was calculated based on the listed equations (details in Methods). Eq, equation; min, minutes; s, seconds.

Source data

Results

Microvascular volumetric pulsatility in pial mater, GM and WM

We first segmented the GM into six equi-volume layers (layers 1–6 from superficial to deep), one layer of superficial WM and an additional pial mater layer on the cortical surface. Due to its thinness, the pial mater layer here also includes adjacent subpial and subarachnoid spaces. Segmentation was derived from T1-weighted MRI (Extended Data Fig. 1a). The deep WM region of interest (ROI) was obtained by subtracting the superficial WM from the total WM. This enabled calculation of laminar profiles of VASO and ASL signals in pial mater, superficial and deep WM and six cortical layers (collectively referred to as cortical laminae). We calculated the laminar profile of cerebral blood flow (CBF) from ASL data for each participant. Extended Data Fig. 2a shows a representative CBF map, and Extended Data Fig. 2c shows a zoomed-in view of the superimposed segmentation, demonstrating the high quality of iso-1.25 mm CBF maps acquired using our three-dimensional turbo-FLASH (TFL) pseudo-continuous ASL (pCASL) at 7 T (ref. ²⁵). Figure 2a illustrates the average laminar CBF and calculated CBV₀ profiles of 11 young participants. A one-way repeated-measures ANOVA revealed a significant main effect of lamina (F_8,80 = 12.34, P < 0.001, ${\eta }_{p}^{2}$ = 0.523 in CBF; F_8,80 = 131.80, P < 0.001, ${\eta }_{p}^{2}$ = 0.921 in CBV₀), indicating differences in resting CBF and CBV₀ across cortical layers. The CBF profiles exhibit the highest values in the middle and superficial layers of the GM, aligning with the vascular density profiles reported by a previous study²⁶. Based on equation (5) in Methods, we estimated the CBV changes across a cardiac cycle (ΔCBV = CBV_max − CBV_min) from VASO signals and plotted the laminar profile of ΔCBV in Fig. 2b. A one-way repeated-measures ANOVA revealed a significant main effect of lamina (F_8,80 = 28.66, P < 0.001, ${\eta }_{p}^{2}$ = 0.718), indicating that ΔCBV progressively increases from the WM to superficial layers of the GM and pial mater. This observation aligns with the reported laminar profile of baseline CBV by Akbari et al.²⁷. Using the method outlined in Fig. 1 (refer to Methods for details), we derived the laminar profile of the mvPI (Fig. 2c) in the whole imaging slab. A one-way repeated-measures ANOVA revealed a significant main effect of lamina (F_8,80 = 19.66, P < 0.001, ${\eta }_{p}^{2}$ = 0.636), illustrating lamina-specific differences in mvPI. Extended Data Fig. 3a shows a matrix to quantify the differences in mvPI values across various cortical laminae. In conjunction with Fig. 2c, we found that mvPI has the highest value in cortical pial mater, which is significantly higher than all other laminae (Bonferroni correction, factor = 36, all corrected P < 0.001). In general, the cortical pial mater exhibits the highest mvPI value (0.18 ± 0.04), and the mvPI gradually decreases as the pial arteries enter the GM, reaching its lowest point in the deep WM.

Fig. 2: Pulsatility maps and laminar results of CBF, CBV₀, ΔCBV and mvPI.

Laminar profiles of CBF value (a), calculated CBV₀ (a), ΔCBV (b) and mvPI (c) show the laminar changes of these parameters. The light blue shaded area indicates the WM area, and the light green shaded area indicates the pial mater area. WMd and WMs on the x axis represent the deep WM and superficial WM, respectively. The values on the x axis from 1 to 0 represent the GM from deep to superficial layers. Statistical significance was assessed by a two-sided bootstrap test (10,000 resamples) with Bonferroni correction (factor = 36). Exact P values are shown in the figure (n = 11 participants, biological replicates). Error bars indicate s.e.m. across participants. Extended Data Fig. 3a shows all comparison statistical results between laminae. d, Laminar mvPI profiles in ACA (blue), MCA (red) and PCA (green) vascular territories. The pink line shows the clustered permutation test P value between ACA and MCA. The yellow line shows the clustered permutation test P value between ACA and PCA. e, mvPI maps in a representative participant. f, Partially zoomed view of the mvPI map. NS, not significant.

Source data

Microvascular volumetric pulsatility in vascular territories

Given the different vascular compliance²⁸ and supratentorial supply regions²⁹ of the anterior cerebral artery (ACA), middle cerebral artery (MCA) and posterior cerebral artery (PCA), we compared average mvPI across these territories using a published vascular territory atlas³⁰ (Extended Data Fig. 1b). Figure 2d shows laminar mvPI profiles for ACA, MCA and PCA. Two-way repeated-measures ANOVA revealed significant main effects of lamina (F_8,80 = 53.14, P < 0.001, ${\eta }_{p}^{2}$ = 0.842) and territory (F_2,20 = 9.84, P < 0.001, ${\eta }_{p}^{2}$ = 0.496), plus a significant interaction (F_16,160 = 10.50, P < 0.001, ${\eta }_{p}^{2}$ = 0.512), indicating distinct mvPI distributions among territories. From WM to middle and superficial GM, mvPI in MCA was significantly higher than in ACA (Bonferroni factor = 9, layer 3 → superficial WM, P < 0.001; clustered permutation P = 0.006) and PCA (P < 0.001; permutation P = 0.010). Across all territories, cortical pial mater had the highest mvPI, with steeper declines in ACA and PCA than in MCA as arteries penetrated deeper.

Microvascular volumetric pulsatility maps

Voxel-wise mvPI maps were generated with the aforementioned method. Figure 2e displays the mvPI map of a representative participant. It indicates that mvPI is not uniformly distributed across the cortex, with regional high mvPI values observed in the WM. Figure 2f offers a detailed view of the mvPI map in a partially zoomed view of Fig. 2e, highlighting the distribution patterns within the GM. The arrows in this figure pinpoint areas with notably high mvPI values located in the superficial GM and pial mater. This observation aligns with the finding in the laminar profile, indicating obviously higher mvPI values in the pial mater and superficial GM compared to the deep and middle GM. Extended Data Fig. 4a shows the mvPI maps of other participants, demonstrating the high quality of mvPI maps calculated by our proposed method.

Reliability of mvPI measurements

The results presented above demonstrate a proof of concept of our method for mapping mvPI. We then investigated the reliability of our method through three distinct approaches. First, we conducted a test–retest study on six participants with a long interval (details in Extended Data Table 1; three are approximately 3 months, and another three are 5–8 months). Second, we employed a non-parametric reliability test to evaluate the robustness of our collected data. Lastly, we generated simulated data to analyze the influence of various parameters on the outcomes.

Figure 3a presents the mvPI maps from two scan sessions, conducted approximately 3 months apart, for three participants. Similarly, Extended Data Fig. 4a illustrates mvPI maps from two scans for an additional three participants, obtained over a longer interval. These images reveal that the mvPI distribution patterns remain highly consistent across the two scans for the same individual yet exhibit marked differences across different participants. To accurately quantify the differences between the images, we calculated the mean squared error (MSE) values for each pair within the six maps. The MSE value calculated between the two mvPI maps provides a comprehensive measure of the differences in both the amplitude and spatial distribution of mvPI. Figure 3b presents the MSE matrix, strengthening the observed patterns. Figure 3c further visualizes these values, demonstrating a significant distinction in the MSE distribution between maps of the same participant versus those from different participants (P = 0.0001). This finding underscores the reproducibility of the mvPI maps despite the relatively long interval between the two scans. Moreover, the capability to visualize individual mvPI maps in each participant enhances the potential for future clinical applications.

Fig. 3: The results of test–retest, reliability test and simulations.

a, mvPI maps in three young participants and two different scans. b, MSE value matrix calculated by test and retest mvPI maps from six young participants. Red arrows highlight the MSE values from different scanning sessions of the same participant. c, Distribution of the MSE values in b. d, The results of the reliability test for three young participants with scan data from two sessions. The red line and red shading represent the mean laminar profile of ΔVASO and its 95% confidence interval, respectively, calculated from random data. The black line shows the laminar profile of the ΔVASO calculated for each participant. e, Average laminar profile of RI across the young participants. Statistical significance for all panels was assessed using a two-sided bootstrap test (10,000 resamples), with Bonferroni correction for multiple comparisons (factor = 9). Exact P values are shown in the figure (n = 11 participants, biological replicates). The top line shows the clustered permutation test result between RI and chance level. Error bars indicate s.e.m. across participants. f, Laminar RI profiles in ACA (blue), MCA (red) and PCA (green). The top lines show the clustered permutation test results between RI and chance level in ACA (blue), MCA (red) and PCA (green). g, Results of simulations. It shows the impact of various univariate changes (variations from left to right: tSNR, tissue volume, actual PI, CBV₀ and breathing effect) on the calculated PI (blue) and RI (red). h, Baseline changes by breathing effect in simulation. It shows the chance level (blue) and 95% confidence interval upper (red).

Source data

As a second reliability check, we performed a non-parametric test by randomly shuffling the VASO time series to remove cardiac synchronization and computing ΔVASO for each shuffle. Repeating this 10,000 times produced a chance-level distribution. Figure 3d and Extended Data Fig. 5a show results for 11 young participants (six with test–retest). The measured ΔVASO (black line) exceeded the 95% confidence interval of the random distribution (red shading) at most layers, indicating high reliability and repeatability, with pial mater and superficial GM more reliable than deeper GM and WM. Reliability in WM was less stable. The temporal signal-to-noise ratio (tSNR) profile of VASO (Extended Data Fig. 5c) inversely matched the chance-level envelope, supporting the accuracy of the baseline estimation. We defined a reliability index (RI; equation (7)) to quantify detection above chance. RI > 1 indicates reliable results. Figure 3e shows that all laminae except deep WM had ΔVASO significantly above chance (Bonferroni corrected, P < 0.001; permutation P < 0.001). Deep WM pulsation was too weak for reliable mvPI estimation. Across vascular territories (Fig. 3f), ACA and PCA pulsatility was reliably detected only in pial mater and superficial GM, whereas MCA showed detectable pulsatility in all laminae except deep WM, consistent with mvPI being greater in MCA than ACA/PCA from middle GM to superficial WM. These results confirm that mvPI reflects genuine CBV changes from vascular pulsation, with clear laminar-specific and territory-specific differences. The RI metric can further serve as a quality control tool for clinical applications.

Considering multiple factors (tSNR, ROI size, breathing effect, CBV₀, etc.) that may influence results, we generated simulated data based on parameters from our dataset (Extended Data Fig. 6a). Figure 3g shows the effects of univariate changes in voxel tSNR, tissue volume, actual mvPI, CBV₀ and breathing effect on calculated mvPI and RI. When voxel tSNR was greater than 5, mvPI and RI remained stable, approximating the set value. In our ROIs, tSNR generally exceeded 7 (Extended Data Fig. 6e), indicating minimal impact. Tissue volumes of more than 2 ml also yielded stable estimates; actual ROI volumes generally exceeded 5 ml (Extended Data Fig. 6d). Varying the true mvPI confirmed accurate estimation: RI increased with mvPI and plateaued at mvPI of approximately 0.2, consistent with our observed maximum (Fig. 2c). CBV₀ changes of more than 4% in GM/pial mater had little effect (Extended Data Fig. 6c), whereas 3–4% changes in WM influenced results (Extended Data Fig. 6b). Breathing effects did not affect mvPI, likely due to cardiac–respiratory independence, but reduced RI by raising the chance-level baseline (Fig. 3h). Overall, simulations support the reliability of our measurements and reveal genuine differences in microvascular pulsatility across vascular territories and laminae.

mvPI changes with aging

A decreasing CBF, lower metabolic rates of glucose and oxygen and a compromised structural integrity of the cerebral vasculature, especially the microvasculatures, are representative degenerative features of the vascular system of the aging brain³¹. To examine age effects on mvPI, we analyzed data from 12 older participants using the same procedures. The laminar mvPI profile (Fig. 4a, red) and inter-lamina difference matrix (Extended Data Fig. 3b) were compared with young participants (Fig. 4a, blue). Two-way ANOVA revealed significant main effects of laminae (F_8,160 = 21.21, P < 0.001, ${\eta }_{p}^{2}$ = 0.485) and age (F_1,20 = 24.06, P < 0.001, ${\eta }_{p}^{2}$ = 0.118). Older participants showed significantly higher mvPI in deep WM (Bonferroni, factor = 9, P = 0.006) and a markedly different distribution pattern (Fig. 4a and Extended Data Fig. 3a,b; Bonferroni, factor = 36). Young participants exhibited a monotonic mvPI decrease from pial mater to deep WM, whereas older participants showed a decline from pial mater to deep GM followed by a plateau. Extended Data Fig. 3d,e display CBV₀ and ΔVASO profiles for both groups. No significant CBV₀ difference was found in deep WM, but ΔVASO differed significantly (Bonferroni, factor = 9, P = 0.003), with a smaller P value than mvPI. Combined with the absence of CBV₀ differences, this indicates that the deep WM mvPI increase in older adults mainly reflects VASO signal changes from microvascular pulsation rather than CBF differences.

Fig. 4: Results of mvPI and RI in older group and the MSE distribution between mvPI maps of different groups.

a, Average laminar profile of mvPI in the young and older participants. Statistical significance for all panels was assessed using a two-sided bootstrap test (10,000 resamples). Bonferroni correction, factor = 36. Exact P values are shown in the figure (young n = 11 and older n = 11 independent biological replicates). Extended Data Fig. 3a,b shows all comparison statistical results between laminae in young and older. Error bars indicate s.e.m. across participants. b, Average laminar profile of RI in the young and older participants. Bonferroni correction, factor = 9; exact P values are shown in the figure; ***corrected P < 0.001, **corrected P < 0.01 and *corrected P < 0.05. The exact corrected P values for the comparison between young and older participants from deep WM to pia are as follows: <0.001, <0.001, 0.003, 0.017, 0.038, 0.004, 0.006, 0.004 and 0.088. c, MSE distribution in four groups: (1) between two scans of the same participant (red, six MSE points); (2) between different young participants (light blue, 130 MSE points); (3) between young and older participants (dark green, 187 MSE points); and (4) between different older participants (dark blue, 55 MSE points). The shape of each violin represents the kernel density estimation of the data. Embedded box plots indicate the interquartile range (IQR; box limits) and the median (center line). Whiskers represent the data range within 1.5× IQR from the box limits. Exact P values are shown in the figure. Bonferroni correction, factor = 6.

Source data

Figure 4b shows the laminar RI profiles for older (red) and young (blue) groups. In older participants, RI values in all laminae were significantly higher than 1 (Bonferroni factor = 9, all corrected P < 0.001; clustered permutation P < 0.001), indicating reliable detection of CBV changes from vascular pulsation across all laminae. Unlike the young group, where deep WM changes were not reliably detected, older participants showed significant CBV changes in deep WM. RI values were also significantly higher in older participants for both WM (deep and superficial WM, corrected P < 0.001) and GM (deep to superficial layers; corrected P = 0.003, 0.016, 0.034, 0.005, 0.009 and 0.006). Extended Data Fig. 7 shows mvPI and RI profiles by vascular territory, demonstrating that, across all territories, older participants had higher mvPI and RI in deep WM supplied by distal vasculature. In summary, older participants exhibited significantly higher mvPI and greater reliability in overall cortex, particularly in deep WM, compared to young participants. This observation aligns with previous studies that assessed the PI between age groups, whether based on blood flow velocity in larger arteries of humans³² or vascular volumetric changes in rats¹⁸.

Many studies identified hypertension, an age-associated chronic condition, as a major contributor to elevated cerebrovascular pulsatility¹⁸. In our cohort of older participants, half had a diagnosis of hypertension, whereas the other half did not (Extended Data Table 1). To investigate the potential contribution of hypertension to age-related changes in mvPI, we stratified the older participants into hypertensive and normotensive subgroups. As none of the young participants had hypertension, mvPI values in deep WM were compared across three groups: young adults, normotensive older adults and hypertensive older adults (Extended Data Fig. 8a). A one-way ANOVA revealed a significant group effect (F_2,19 = 9.64, P = 0.001, ${\eta }_{p}^{2}$ = 0.504), indicating substantial differences in mvPI across groups. Post hoc comparisons showed that the hypertensive older group had significantly higher mvPI values than both the young group (uncorrected P < 0.001) and the normotensive older group (uncorrected P = 0.007). The normotensive older group also exhibited a trend toward increased mvPI relative to the young group (uncorrected P = 0.056), although this difference did not reach statistical significance. These findings suggest that both hypertension and aging may contribute to increased microvascular pulsatility in deep WM. Although the limited sample size constrains the statistical power of subgroup analyses and precludes definitive conclusions, the significant ANOVA results underscore the sensitivity of our method in detecting mvPI alterations associated with vascular risk factors.

Extended Data Fig. 4b shows mvPI maps of all older participants, which appear higher in WM compared to the young group (Extended Data Fig. 4a). This aligns with the greater laminar mvPI and RI values in older participants (Fig. 4a,b; two-way ANOVA, age main effect P < 0.001 for both mvPI and RI). Previous studies suggested that cerebral vascular variability may increase with age^33,34, and mvPI changes may contribute. We calculated the MSE between all mvPI maps to quantify differences in amplitude and distribution, grouping results into (1) two scans of the same participant (0.072 ± 0.024), (2) different young participants (0.099 ± 0.023), (3) young versus older (0.155 ± 0.070) and (4) different older participants (0.195 ± 0.079) (Fig. 4c; Bonferroni factor = 6). MSE in group 1 was significantly lower than in other groups, indicating high repeatability and that WM mvPI heterogeneity reflects physiological, not technical, differences. Group 2 MSE was lower than group 4, supporting increased mvPI variability with age. Groups 3 and 4 had significantly higher mean and s.d. than group 2, and group 3 was lower than group 4, indicating that the variability increase is mainly due to growing differences among older participants rather than a uniform mvPI elevation.

Pulsatility of velocity and volume in the major cerebral arteries via 4D-flow PC-MRI

Finally, we used 4D-flow PC-MRI to examine the pulsations within the major cerebroarteries and their relationship with the mvPI. Previous studies demonstrated the feasibility of using 4D-flow to simultaneously measure the PI of velocity and volume at different locations of blood vessels^35,36,37. As shown in Fig. 5a, we selected two ICA sites, one ACA site, three MCA sites and two PCA sites for pulsatility measurements. Figure 5b shows the cardiac phase profiles of velocity and cross-sectional area from ICA for a representative participant; similar profiles were obtained for each ROI. Using equations (8) and (9), we calculated velocity (vePI) and volumetric (voPI) PIs for both age groups (Fig. 5c and Extended Data Fig. 9b).

Fig. 5: vePI and voPI calculation and results from 4D-flow dataset.

a, Schematic diagram of the ROI selection position of a typical subject in the circle of Willis. The lower left corner is the sagittal plane image within the black frame. b, The cardiac phase profile of velocity and cross-sectional area obtained by ICA from a random participant. c, The vePI and voPI results in both age groups within all MCA ROIs. The top image is the results of vePI, and the bottom one is the results of voPI. Red is for older participants, and blue is for young participants. The circles are the results for each participant, and the horizontal line is the average of this group of data. Statistical significance was assessed using a two-sided bootstrap test (10,000 resamples). Bonferroni correction, factor = 3. Exact P values are shown in the figure (young n = 9 and older n = 12 independent biological replicates). Error bars indicate s.e.m. across participants. d, Relationship between mvPI in deep WM and vePI in ICA and MCA1. The shading is the 95% confidence interval, and dots represent individual participants. n = 20 independent biological replicates. R² indicates how much of the variance in the dependent variable is explained by the independent variable. e, Changes of vPI (including mvPI and voPI) as the cerebral vasculature in MCA territory. The blue and red curves represent the results for young and older participants, respectively. The light blue shaded area indicates the WM area, and the light green shaded area indicates the pial mater area. WMd and WMs on the x axis represent the deep WM and superficial WM. The values on the x axis from 1 to 0 represent the GM from deep to superficial. M1, M2 and M3 represent the MCA1, MCA2 and MCA3 segment. Young n = 9 and older n = 11 independent biological replicates. Error bars indicate s.e.m. across participants.

Source data

We focused on MCA results, as our field of view (FOV) primarily covered MCA regions, whereas coverage in other vascular territories was limited. The results for other regions are provided in Extended Data Fig. 9b. Figure 5c shows mean vePI and voPI for all MCA ROIs. For vePI, two-way ANOVA (vascular segment × age) revealed a main effect of segment (F_2,56 = 25.74, P < 0.001, ${\eta }_{p}^{2}$ = 0.475), with M3 significantly lower than M1 (Bonferroni factor = 6, P < 0.001 in both groups), indicating decreased vePI in distal arteries, consistent with previous work^36,37. For voPI, two-way ANOVA showed a main effect of segment (F_2,56 = 15.81, P < 0.001, ηp² = 0.357), with M3 significantly greater than M1 (Bonferroni factor = 6, P < 0.001 in both groups), opposite to the vePI trend and matching previous findings³⁶. To investigate the age effects on vePI and voPI, we conducted a two-way ANOVA for vePI that revealed a significant main effect of age (F_1,56 = 34.02, P < 0.001, ${\eta }_{p}^{2}$ = 0.374) in MCA ROIs, with vePI being significantly higher in the older group compared to the young group (Bonferroni factor = 9, as shown in Fig. 5c). This finding indicates that vePI increases with age, consistent with previous studies^32,38,39 assessing vascular health. In contrast, voPI showed no significant age effect (F_1,56 = 2.36, P = 0.130). Over the past several years, vePI in the ICA and M1 segment of MCA has frequently been used as an indicator to assess vascular health^3,4,35,40, and vePI values have been shown to be associated with SVD^1,2,3,4 and cognitive impairment in Alzheimer’s disease^5,6,7,8. Next, we conducted a correlation analysis between the mvPI in deep WM calculated using the proposed method and the vePI in the ICA and M1 calculated using 4D-flow (Fig. 5d). Significant correlations exist between the mvPI in deep WM and vePI in the ICA and M1 (Pearson correlation: r = 0.5173, P = 0.0195 with ICA; r = 0.5618, P = 0.0099 with M1; n = 20, degrees of freedom = 18, two-sided test).

To characterize the dynamic changes in volumetric pulsatility along the cerebrovascular network, we combined the mvPI results in the MCA territory with voPI measurements from the MCA by 4D-flow (Fig. 5e). We found that the vPI (including both mvPI and voPI) gradually increased from proximal to distal segments in large arteries, reaching a maximum value in pial arteries within the pial mater on cortical surface. Subsequently, the mvPI decreased with increasing cortical depth, reaching a minimum in the deep WM. However, there was a significant overall increase in mvPI with aging (two-way ANOVA showed a significant main effect of age, F_1,20 = 24.06, P < 0.001, ${\eta }_{p}^{2}$ = 0.118). In particular, there was a significant increase in mvPI with aging within the deep WM compared to other locations (Bonferroni factor = 9, corrected P = 0.018).

Discussion

Using our innovative method, we obtained the distribution of mvPI in both GM and WM. Simultaneously, we employed the conventional 4D-flow MRI to determine pulsatility of blood flow velocity and blood volume along the large arteries. By integrating these findings, we characterized the dynamic changes in volumetric pulsatility throughout the entire cerebrovascular tree from the large arteries to the arterial terminals of the WM. We found that arterial volumetric pulsatility peaks in the pial mater on the cortical surface, and the volumetric PI gradually decreases in both upstream arteries and downstream microvasculature. Compared to young participants, the mvPI in the GM and WM of the older participants was overall substantially increased, especially in the WM.

Several previous studies investigated the vascular origins of the VASO functional MRI (fMRI) signal⁴¹. VASO is thought to be sensitive primarily to CBV changes in pial arteries⁴², larger diving arterioles⁴³, arterioles⁴³ and capillaries⁴⁴, whereas contributions from venules are expected to be weak⁴⁵. Thereby, our method predominantly reflects the pulsatility of small arteries and capillaries, including pial arteries and the arteries smaller than them.

We produced a probability density distribution map of high mvPI (Extended Data Fig. 4c; details in Methods). We observed that regions with high mvPI frequently localize to watershed areas between different vascular territories in the WM. These regions are well known to be high-risk zones for various neurological disorders⁴⁶. Anatomically, the vasculature in these areas primarily consists of distal arteries and capillaries, which supports the plausibility of elevated mvPI in these locations. Furthermore, the probability density map of high mvPI shows substantial overlap with WM hyperintensity probability density map⁴⁷, suggesting a potential link. Notably, we also found that, in the frontal lobe, high mvPI tends to occur more frequently on the right side than on the left. This right-lateralized distribution is consistent with previous findings showing age-related reductions in CBF predominantly in the right frontal lobe⁴⁸, suggesting a possible underlying relationship between these parameters.

The reliability of our method is strongly supported by rigorous validation processes. We employed non-parametric testing to confirm that our results are significantly higher than chance-level distributions, ensuring that they stem from genuine volumetric pulsatility changes rather than noise. Consistency across multiple scans of the same participants further demonstrates the stability of our method, with similar mvPI distribution patterns observed despite relatively long time gaps (3–8 months). The strong correlation between the mvPI results by our method and the vePI results obtained using 4D-flow further validates the reliability of our method. Additionally, our findings align with previous in vivo studies, particularly the high mvPI in pial mater^{9,14,15,18,49} and significantly increased mvPI in WM microvasculature with aging or disease in animal models¹⁸. These factors collectively affirm the reliability and potential of our method as a valuable imaging tool for assessing microvascular health in a living human brain. However, our method has a limitation in that we cannot determine a specific CBV₀ value for each participant and must rely on an approximate value (0.055) derived from previous literature. We attempted to set a different CBV₀ value (0.04) and found that the results remained largely consistent with those of the present study (Extended Data Fig. 8b). Although the calculated mvPI values differed, the statistical analysis and distribution patterns were essentially unchanged.

Our method shows strong potential for both research and clinical use, enabling accurate imaging of the cerebral microvascular system and capturing volumetric pulsatility in GM and WM. The mvPI proved sensitive to aging and vascular risk factors such as hypertension, consistent with a preclinical study in hypertensive mice¹⁸. This approach supports studies of neurodegenerative diseases and the cerebral glymphatic system, with implications for early diagnosis, monitoring and mechanistic exploration. For clinical applications, long scan times may be impractical. Because the deep WM mvPI could serve as a primary indicator, using lower-resolution sequences on 3 T or 7 T MRI to improve SNR may help shorten scans. Additionally, integrating a recent velocity selective ASL method⁵⁰ for measuring CBF pulsatility with our approach could enable microvascular volume and blood flow pulsation maps within a clinically feasible session. In this study, we used the method to preliminarily investigate age effects on mvPI and its distribution. Next, combined with past research findings, we aim to propose a potential mechanism explaining the impact of age and vascular risks on microvascular pulsatility, thereby highlighting the feasibility and potential of our innovative method for advancing future research in this area.

Cerebral arterioles are key regulators of arterial pressure, showing a marked drop compared to large arteries^51,52. Their small diameter greatly increases resistance, as described by the Poiseuille–Hagen formula (resistance ∝1 / r⁴), and their extensive branching further adds to this effect. As blood flows from large arteries into arterioles, the total cross-sectional area increases, lowering velocity and pressure (continuity equation A₁v₁ = A₂v₂). Smooth muscle cells in the arteriolar wall actively adjust diameter to regulate local blood flow and systemic pressure^53,54. Large arteries, with their low resistance, cannot substantially reduce pulse pressure⁵¹, whereas arterioles dissipate it effectively through higher resistance and elastic wall expansion, distributing flow and protecting downstream capillaries^51,55. On the brain’s surface, leptomeningeal and pial arteries possess an internal elastic lamina, basement membrane and smooth muscle layers^14,56,57. Once they penetrate the cortex, they lose the elastic lamina and adventitia, becoming arterioles. Surface arterioles have densely arranged smooth muscle cells and abundant elastic fibers, enabling them to withstand higher pressures and dynamic changes^58,59. By contrast, intracortical arterioles have fewer muscle cells, less dense arrangement and fewer elastic fibers, resulting in softer walls. Their main role is to supply neurons and glia within the GM, participate in local flow regulation and match metabolic demands. These anatomical differences have direct hemodynamic consequences. By the time blood reaches intracortical arterioles, much of the arterial and pulse pressure has been attenuated, reducing the need for further dissipation. Consequently, volumetric pulsatility is stronger in surface arterioles (Fig. 6). Unlike vPI, the Bernoulli equation (v² ∝ P) shows that vePI depends on both pulse pressure and mean velocity; as both decrease distally, vePI correspondingly declines (Fig. 6). In summary, from hydromechanical and hemodynamic perspectives, cortical surface arteries exhibit the strongest volumetric pulsatility in the brain, in agreement with our observation of maximal pulsatility in the pial mater.

Fig. 6: Hypothesis model on physiological mechanisms of laminar changes in arterial pulsatility.

a, Potential physiological mechanisms of arterial pulsatility changes with laminae in normal or abnormal brain. The blue box at the top of the image illustrates the physiological mechanisms underlying the changes in vPI and vePI across arteries in young, healthy brains. Upward-pointing arrows indicate that these parameters increase along the arterial transmission pathway, whereas downward-pointing arrows signify a decrease. Conversely, the red box at the bottom of the image depicts the mechanisms of abnormal vPI and vePI in elderly or pathological brains. In this context, red arrows indicate an increase in parameters relative to healthy brains, whereas blue arrows denote a decrease. b, The laminar profiles distribution of arterial total volume, blood pressure, predicated vPI (including mvPI and voPI) and vePI in young and elderly individuals predicted by the hypothetical model.

Source data

In addition to reducing pulse pressure and protecting small vessels, cerebral arterial volumetric pulsation is a key driver of fluid circulation and waste clearance^{9,14,49,60,61}. Experiments injecting molecules of various sizes into the rat caudate nucleus and cat CSF showed that interstitial solutes and CSF drain via the PVS and have identical half-lives, indicating bulk flow rather than diffusion^10,62,63,64. Louveau et al.¹⁵ proposed that the PVS acts as a pseudo-lymphatic system, delivering fresh CSF to the interstitium and removing metabolic waste. Critically, cyclical vessel wall motion from arterial pulsation drives CSF movement¹⁸. Thus, the strong cortical surface arterial pulsation detected by our method can propel large volumes of fresh CSF into the brain parenchyma, enhancing fluid circulation and waste removal.

Aging brains show vascular degeneration, including reduced CBF, lower glucose and oxygen metabolism and structural compromise, especially in microvessels³¹. CBF reduction with aging likely reflects a shift toward constrictor responses, with impaired vasodilation or enhanced vasoconstriction^65,66,67, compounded by age-related loss of perivascular innervation^68,69. Microvascular density declines in both humans⁷⁰ and animals^71,72,73 with aging, accompanied by perivascular collagen deposition and basement membrane thickening^74,75,76,77, reducing arteriole volume and branching and limiting their ability to dissipate arterial and pulse pressures⁷⁸. Residual pressure propagates distally to distal arteries in WM, which cannot release it, mechanically increasing volumetric pulsation and producing strong wave reflections^67,79. The effective reflection distance of the reflected wave increases in older participants compared to young participants⁷⁹, indicating that the main reflection points of the reflected wave move distally (that is, the main source of wave reflection changed from various arterial branches to the reflection of the WM artery at the end of the artery) and support our inference according to fluid mechanics. Although wave reflection–pulsation interactions are complex, our data and rat studies¹⁸ suggest that this process enhances arterial volumetric pulsation, especially near terminal arteries. Because vascular volume changes are influenced by both direct cardiac input and reflections, the vascular wall’s time–change curve is altered, disrupting synchrony between pulsation and CSF flow. This can cause CSF reflux and markedly slow flow rate¹⁸. In summary, aging leads to markedly increased volumetric pulsation in WM arteries, with only slight increases in other arteries (Fig. 6).

In addition to aging, conditions such as hypertension and arteriosclerosis can abnormally increase arterial volumetric pulsation^78,80. Prolonged exposure to elevated pulsation raises the risk of vascular degeneration, including fibrosis, particularly in small WM vessels⁸¹, leading to SVD and cerebral microbleeds^81,82,83. If unchecked, these changes increase the likelihood of stroke, dementia, cognitive decline and related disorders^82,83. High pulsatility also slows CSF flow¹⁸ and impairs waste clearance, promoting the accumulation of metabolic byproducts such as β-amyloid, which further obstruct clearance and may contribute to Alzheimer’s disease^82,83,84. Therefore, detailed assessment of vascular volumetric pulsation can aid in evaluating vascular disorders such as SVD and cognitive diseases such as Alzheimer’s, providing broader insight into their underlying causes.

In summary, our method combining high-resolution VASO and three-dimensional TFL-pCASL MRI offers in vivo fine-grained measurement of microvascular volumetric pulsatility in the human brain. Our findings of the highest mvPI in the pial mater on the cortical surface and a significantly increased mvPI in the WM of older participants compared to younger ones have implications for cerebral microvascular health and its changes with aging and neurodegenerative diseases. Our technology offers a valuable imaging tool for understanding hydromechanical effects on cerebral microvasculature, fluid circulation and brain health.

Methods

Participants

Eleven young volunteers (mean age = 28.4 years, s.d. = 5.8 years, five females) and 12 older volunteers (mean age = 60.2 years, s.d. = 6.8 years, seven females) participated in this study. All participants in this study were healthy without known neurological or psychiatric disorders or major systemic diseases and refrained from caffeine 3 hours before the scan. All participants provided written informed consent according to a protocol approved by the institutional review board of the University of Southern California and were paid for attendance. Head motion was minimized by placing cushions on top and on two sides of the head. All participants underwent pCASL and VASO scans with recording photoplethysmogram (PPG) signals on the finger by BioPac. Six of the young participants underwent repeated VASO scans on two different days with long interval (details in Extended Data Table 1) for test and retest. Nine of the young participants and all the older participants were scanned with a 4D-flow sequence.

MRI data acquisition

MRI data were acquired on the investigational pTx part of a 7 T Terra system (Siemens Medical Systems) with an 8Tx/32Rx head coil (Nova Medical). Anatomical images were acquired with a T1-weighted MP2RAGE sequence (0.7-mm isotropic voxels, 224 sagittal slices, FOV = 224 × 224 mm², TR = 4,500 ms, TE = 3.43 ms, TI1 = 1,000 ms, flip angle = 4°, TI2 = 3,200 ms, flip angle = 5°, bandwidth = 200 Hz/Px, slice partial Fourier = 6/8 and GRAPPA = 3).

Functional images were acquired for resting states with a slice-selective slab-inversion (SS-SI) VASO^24,85,86,87 sequence (1-mm isotropic voxels, 24 axial slices, TI1/TI2/shotTR/TE = 1,367.4/2,512.2/47.7/16.2 ms, partial Fourier factor = 6/8, variable flip angle scheme with reference flip angle = 33°, three-fold GRAPPA acceleration in z direction with a CAIPIRINHA shift of 1, bandwidth = 1,644 Hz/Px, FOV = 160 mm, matrix size = 160 × 160 and acquisition time approximately 30 minutes), a three-dimensional TFL-pCASL^25,88 sequence (1.25-mm isotropic voxels, 80 axial slices, FOV = 224 × 200 × 100 mm³, flip angle = 8°, TE = 2.5 ms, TR = 7 s, labeling duration = 1,500 ms, post-labeling delay = 1,500 ms, 10 segments, Poisson disc undersampling with R = 4, acquisition time approximately 12 minutes and Compressed Sensing reconstruction) and a 4D-flow PC-MRI sequence (0.8-mm isotropic voxels, 40 axial slices, FOV = 200 × 200 × 32 mm³, reconstructed cardiac phases: 12, flip angle = 15°, TR/TE = 64.32/2.87 ms, SENSE: 3, RL, AP and FH velocity encoding: 100 cm s⁻¹, acquisition time approximately 8 minutes and Compressed Sensing reconstruction).

MRI data preprocessing

MRI data were analyzed using AFNI⁸⁹, FreeSurfer⁹⁰, ANTs⁹¹, SPM12 (ref. ⁹²), the mripy package (https://github.com/herrlich10/mripy) and MATLAB (R2021a). The preprocessing of the VASO data includes removing spikes in time series, linear motion correction and T1-weighted anatomical image registration to functional volumes. ASL data were reconstructed, and CBF values were calculated using the method in previous studies^25,88. We estimated a 12-parameter linear transformation from the anatomical volume to an MNI space template by nonlinear transformation with ANTs. Subsequently, we can use the vascular territory atlas³⁰ in the standard brain template and transformed into the individual space of each participant using nonlinear transformation parameters. The 4D-flow MRI data preprocessing involved background phase correction and eddy current effects⁹³. Velocity aliasing was corrected using phase unwrapping techniques⁹⁴. Finally, the three-dimensional vector of blood flow velocity was computed for each voxel.

Cortical layers definition

VASO and ASL images were separately upsampled to a finer grid of 0.5 mm³ and 0.625 mm³ isotropic spatial resolution to avoid singularities at the edges in angular voxel space using the ‘3dresample’ program in AFNI with the nearest neighbor interpolation method. The T1-weighted MP2RAGE anatomical volume was segmented into WM, GM and CSF using FreeSurfer’s automated procedure and its high-resolution option. A boundary-based algorithm was separately used to co-register structural MP2RAGE images to VASO and ASL control images⁹⁵. Eight cortical layers (one WM, six GM and one pial mater) were calculated using AFNI/SUMA and custom Python codes in the mripy package with the equi-volume layering approach⁹⁶. Although the cortical depths of different brain regions vary (from approximately 4-mm thickness of the prefrontal cortex to approximately 2-mm thickness of the primary sensory cortex) and the voxel size is not large enough to clearly distinguish the cortical layers, a large number of voxels randomly sampled at different cortical depth allowed us to derive a reliable continuous laminar profile of neural activity²⁴, as shown in Extended Data Fig. 1a.

PI calculation in VASO

Previous studies generally used (V_max − V_min) / V_mean (V is velocity or blood flow) as the PI calculation equation^35,97. To estimate the volume change of microvasculature across the cardiac cycle, we used the following formula to calculate the mvPI:

$${\mathrm{mvPI}}=\frac{{{\mathrm{CBV}}}_{\max }-{{\mathrm{CBV}}}_{\min }}{{{\mathrm{CBV}}}_{0}}=\frac{\Delta {\mathrm{CBV}}}{{{\mathrm{CBV}}}_{0}}$$

(1)

where CBV_max, CBV_min and CBV₀ are the CBV values when the blood vessels expand to their maximum volume, contract to their minimum volume and averaged over the entire cardiac cycle, respectively. ΔCBV is absolute CBV changes from CBV_min to CBV_max. According to the principle of VASO sequence⁹⁸:

$${\mathrm{VASO}}=M(1-{\mathrm{CBV}})$$

(2)

Equation (2) can be converted into the following:

$${\mathrm{CBV}}=1-\frac{{\mathrm{VASO}}}{M}$$

(3)

$$M=\frac{{{\mathrm{VASO}}}_{0}}{1-{{\mathrm{CBV}}}_{0}}$$

(4)

According to equations (3) and (4), we can deduce:

$$\begin{array}{l}\Delta{\mathrm{CBV}}={{\mathrm{CBV}}}_{\max}-{{\mathrm{CBV}}}_{\min}\mathop{\Leftrightarrow}\limits^{\left[3\right]}\,\displaystyle\frac{{{\mathrm{VASO}}}_{\min}-{{\mathrm{VASO}}}_{\max}}{M}\\\qquad\qquad\mathop{\Leftrightarrow}\limits^{\left[4\right]}(1-{{\mathrm{CBV}}}_{0})\displaystyle\frac{{{\mathrm{VASO}}}_{\min}-{{\mathrm{VASO}}}_{\max}}{{{\mathrm{VASO}}}_{0}}\end{array}$$

(5)

where VASO_min is the VASO signal when it reaches CBV_min, and VASO_max is the VASO signal when it reaches CBV_max. Substituting equation (5) into equation (1), we can derive the equation for mvPI:

$${\mathrm{mvPI}}=\left(\frac{1}{{{\mathrm{CBV}}}_{0}}-1\right)\frac{{{\mathrm{VASO}}}_{\min }-{{\mathrm{VASO}}}_{\max }}{{{\mathrm{VASO}}}_{0}}$$

(6)

Figure 1 shows the calculation process of the mvPI using VASO and ASL data, including the following. (1) According to the recorded PPG signal, we defined 10 cardiac phases as shown in Fig. 1a. Then, we can get the cardiac phase at any timepoint like this black line in Fig. 1b, bottom. (2) Because the SS-SI VASO^24,85,86,87 sequence recorded both VASO and blood oxygenation level dependent (BOLD) image series, we recorded the trigger time of all VASO images and BOLD images (blue and red bars in Fig. 1b, bottom). According to the cardiac phase corresponding to the trigger time, we can define the cardiac phase of every VASO/BOLD image (Fig. 1b, top). (3) We rebinned all the VASO and BOLD images with same cardiac phase and averaged it. Then, we got the cardiac phase profiles of both VASO and BOLD images. (4) We divided the VASO signal by the BOLD signal to obtain the cardiac phase profile of the corrected VASO signal (Fig. 1d). From this, we can calculate the value of (VASO_min − VASO_max) / VASO₀ term in equation (6). (5) According to the ASL results, we can calculate the laminar profile of CBF value (Fig. 1c, left). (6) We assumed CBV₀ ~ CBF^0.38 according to the study by Grubb et al.⁹⁹. Assuming average baseline CBV₀ = 0.055 ml ml⁻¹ in the GM¹⁰⁰, we can get the laminar profile of CBV₀ value (Fig. 1c, right). This is the 1 / CBV₀ − 1 term in equation (6). Finally, we can calculate the mvPI value.

Voxel-wise mvPI maps were generated with the aforementioned method. To enhance the SNR, which may be insufficient in a single voxel, we applied intralayer spatial smoothing with a 6-mm full width at half maximum (FWHM) prior to calculation^24,101. Due to the differing voxel sizes between the ASL and VASO sequences, as well as the relative low SNR of individual ASL voxels, we used an average CBF value within each laminar ROI to estimate mvPI.

We generated a high mvPI probability density distribution map by the mvPI maps of all participants projected onto the MNI template (Extended Data Fig. 4c). Due to slight variations in the FOV across participants, we first counted the number of participants covering each voxel, which was defined as the total counts of occurrences for that voxel. We then counted how often each voxel exhibited an mvPI value greater than 0.6—a subjectively defined high threshold based on the overall distribution of mvPI values—to determine the number of high mvPI occurrences at each voxel. The probability density value for each voxel was calculated by dividing the high mvPI counts by the total counts. To minimize the influence of sparse sampling and random variability, we excluded voxels with fewer than 10 total occurrences. This procedure yielded a high mvPI probability density distribution map.

Reliability test

We verified the reliability of the results using a non-parametric test. We selected an ROI to extract the average corrected VASO time series, and the ΔVASO of this ROI with cardiac cycle can be calculated using the calculation method described above equations (1)–(4). Then, we randomly shuffled the order of this time series and performed the calculation again using the same method to get a ΔVASO value. Repeating this step 10,000 times, we can get a ΔVASO value distribution of results representing no vascular pulsatility with a sample size of 10,000. We can get the mean and 95% confidence interval of this distribution, and we can also get the P value that our result satisfies the non-hypothesis that the measurement belongs to this distribution (the number of results in the distribution that are greater than the original ΔVASO divided by the sample size). We further defined an RI according to the following equation:

$${\mathrm{Reliability}}\; {\mathrm{Index}}\left({\mathrm{RI}}\;\right)=\frac{\Delta{\mathrm{VASO}}-{{\mathrm{Distribution}}}_{{\mathrm{ave}}}}{{{\mathrm{Distribution}}}_{{\mathrm{high}}-{\mathrm{CI}}}-{{\mathrm{Distribution}}}_{{\mathrm{ave}}}}$$

(7)

where ΔVASO is the calculated result from the original time series; Distribution_ave is the average value of the random distribution; and Distribution_high-CI is the upper bound of the 95% confidence interval. RI is obtained by dividing the difference between the ΔVASO and the mean of the distribution by the difference between the upper level of the 95% confidence interval and the mean of the distribution. If the RI value is greater than 1, it means the result is reliable and vice versa. As shown in Extended Data Fig. 5, all participants showed reliable measurements (RI > 1) except for S15. Therefore, we used 11 older participants for mvPI analysis. The RI provided an objective and quantitative metric for quality control of our mvPI results.

PI calculation by 4D-flow

We drew several spherical masks with diameters larger than the diameter of the blood vessels at the locations of straight segments of major blood vessels. These included two ICAs, one ACA, three MCAs (M1, M2 and M3 segment) and two PCAs (P1 segment and P2 segment), shown in Fig. 5a. For more fine-grained analyses, we interpolated the resolution to 0.4 mm isotropic for subsequent calculations. Because the selected sites have small vascular curvature, we can calculate the equation of a three-dimensional plane perpendicular to the vessel based on the direction of the average blood flow velocity within the vessel. Using this plane equation, we extracted a cylindrical mask with a thickness of 1 mm, ensuring that the upper and lower planes of the cylinder were perpendicular to the vessel. By calculating the inner product of the blood flow velocity vector of each voxel within the mask and the unit blood flow velocity vector of the vessel, the velocity projection of each voxel in the blood flow direction can be determined. This approach yields a distribution of values across all voxels, comprising two components: (1) a distribution approximating a normal distribution with a mean close to 0, formed by voxels outside the vessel, and (2) a distribution with a higher mean value, formed by voxels within the vessel. To determine whether a voxel belongs to a blood vessel in a specific cardiac phase, we first sorted the distribution values less than 0 and selected the smallest 10% as the lower bound of the one-sided 95% confidence interval for the non-vascular distribution (for example, if there were 100 values less than 0, the 10th smallest value in the sorted list was chosen as the lower bound). The absolute value of this lower bound was then used as the upper limit of the one-sided 95% confidence interval for the non-vascular distribution. This value as a threshold was then applied to determine whether a voxel belongs to a blood vessel during a specific cardiac phase. All voxels with velocity values exceeding the threshold at all cardiac phases were considered part of the vascular mask. Using this mask, we can calculate the cardiac phase profile of blood flow velocity (Fig. 5b, blue line). Due to the laminar flow effect of blood, it is not possible to directly determine the volumetric proportion of the vascular component within a voxel at a given cardiac phase timepoint based solely on the velocity value of the voxel projected in the direction of blood flow. Consistent with a previous study³⁶, for each cardiac phase, we applied half the value of the highest PC-MRI magnitude image as a threshold to identify vascular voxels—that is, a voxel was considered part of the vessel for that cardiac phase only if the vascular volume proportion within the voxel exceeded this threshold. Extended Data Fig. 9a illustrates the volume pulsatility of large and small arteries, respectively. The number of such voxels was then multiplied by the voxel volume (0.4³ mm³) and divided by the cylinder height (1 mm) to calculate the cross-sectional area of the vessel for that cardiac phase. Then, we can get the cardiac phase profile of the arterial cross-sectional area (Fig. 5b, green line). Finally, we can calculate the PI values using the following equations:

$${\mathrm{vePI}}=\frac{{v}_{\max }-{v}_{\min }}{{v}_{{\mathrm{mean}}}}$$

(8)

$${\mathrm{voPI}}=\frac{{S}_{\max }-{S}_{\min }}{{S}_{{\mathrm{mean}}}}$$

(9)

where v means blood flow velocity, and S means vessel cross-sectional area. vePI and voPI present the PI values of the changes in blood flow velocity and arterial volume, respectively.

Calculation of MSE

Image stability assessment such as MSE¹⁰² and peak signal-to-noise ratio (PSNR)¹⁰³ are widely used to measure the quality and stability of image processing. An mvPI map can be calculated based on the data from each visit of any participant. Then, we projected all vmPI maps into MNI space using the nonlinear projection parameters calculated by ANTs. Twelve mvPI maps from six participants, who underwent both test and retest sessions, were selected for analysis (for Fig. 3b,c). We used the common mask intersection of these 12 mvPI maps as the new mask, so all calculations used only the same batch of voxels. Then, we calculated the MSE value between each pair of maps using the following formula:

$${\mathrm{MSE}}=\frac{1}{v}\mathop{\sum }\limits_{i=1}^{v}{[I\left(i\right)-K(i)]}^{2}$$

(10)

where I and K are two mvPI maps, and v is the number of voxels in the mask.

Additionally, we plan to extend this analysis to all participants, including both older and young participants, for the results presented in Fig. 4c. However, due to the small number of remaining voxels from the intersection mask of all mvPI maps, we calculated the MSE between two mvPI maps using just the intersection mask of these two mvPI maps instead of the intersection mask of all mvPI maps.

Simulations

Many previous studies showed that physiological factors, such as cardiac cycle and respiratory cycle, can affect CBV and fMRI signals^104,105,106. Extended Data Fig. 6a displays the simulation process. We assumed that the factors that affect CBV changes are mainly the cardiac phase and respiratory phase and generated CBV time series using sine function. The heart rate was set to 80 ± 10 (mean ± s.d.) times per minute, and the respiratory rate was set to 12 ± 3 times per minute. The PI and RI were both set to 0.2 (based on the results in Fig. 2). In this way, we can obtain the time series of CBV changes with the cardiac phase and respiratory phase, respectively. Then, according to the repeated acquisition time of our sequence (approximately 3.1 s), we resampled and obtained the time series of CBV changes with a length of 600 TRs. We added these two time series together and calculated the time series of the VASO signal without noise according to equation (2). We then added Gaussian noise to the time series of the VASO signal to make the tSNR of a single voxel equal to 7 (the smallest tSNR in Extended Data Fig. 6e). We set the ROI size to 5 ml (that is, 5,000 voxels, the minimum value in Extended Data Fig. 6d) and then used the proposed method to calculate the PI and RI of the generated data. Finally, we can control a single variable to generate a curve showing the effect of a single variable change on PI and RI values, shown in Fig. 3g,h.

Statistical analysis

The statistical tests in this study were performed using MATLAB and JASP software (version 0.16.4). Given the relatively small sample size of our study, we used non-parametric bootstrapping method for the hypothesis tests of cortical laminae-dependent effect and age effect. Cluster-based permutation test was used for the clustered significance of a range of cortical laminae in vmPI and RI. One-way ANOVA and two-way ANOVA were performed for laminar effect or interaction effect of laminae and age or vascular territory. Prior to conducting the ANOVA test, each dataset was evaluated for normality using the Shapiro–Wilk test and for homogeneity of variance using Levene’s test.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.