Three-dimensional assessment of brain arterial compliance: technical development, comparison with aortic pulse wave velocity, and age effect

Abstract

Purpose:

The ability to measure cerebral vascular compliance (VC) is important in the evaluation of vascular diseases. Additionally, quantification of arterial wall pulsation in the brain may be useful for understanding the driving force of the recently discovered glymphatic system. Our goal is to develop an MRI technique to measure VC and arterial wall pulsation in major intracranial vessels.

Methods:

A total of 17 healthy subjects were studied on a 3 Tesla MRI system. The technique, dubbed VaCom-PCASL, uses pseudo-continuous-arterial-spin-labeling (PCASL) to obtain pure blood vessel signal, employs a 3D radial acquisition, and applies a golden-angle-radial-sparse-parallel (GRASP) algorithm for image reconstruction. The k-space data were retrospectively sorted into different cardiac phases. The GRASP algorithm allows the reconstruction of five-dimensional (three spatial dimensions, one control/label dimension, and one cardiac-phase dimension) data simultaneously. The proposed technique was optimized in terms of reconstruction parameters and labeling duration. Intracranial VC was compared with aortic pulse-wave-velocity (PWV) measured with phase-contrast MRI. Age differences in VC were studied.

Results:

VaCom-PCASL using 10 cardiac phases and GRASP sparsity constraints of λ_{label/control}=0.05 and λ_cardiac=0.05 provided the highest contrast-to-noise ratio. A labeling duration of 800 ms was found to yield signals comparable to those of longer duration (p>0.2), whereas 400 ms yielded significant over-estimation (p<0.005). A significant correlation was observed between intracranial VC and aortic PWV (r=−0.73, p=0.038, N=8). Vascular compliance in the older group was lower than that in the younger group.

Conclusion:

VaCom-PCASL MRI represents a promising approach for non-invasive assessment of arterial stiffness and pulsatility.

INTRODUCTION

The compliance of a blood vessel denotes its ability to distend passively in response to an increase in blood pressure. A normal compliance in major arterial vessels, referred to as arterial compliance (AC) in some literatures (1) or vascular compliance (VC) in others (2), is important in attenuating pulsatile blood flow in large arteries into continuous flow in capillaries, thereby protecting vascular bed and parenchyma from injuries due to distal flow pulsatility. A diminishment in vascular compliance, i.e. stiff vessel, has been suggested to be a risk factor for a number of diseases, such as cardiovascular and cerebrovascular diseases, hypertension, diabetes, and vascular dementia (3–10). Furthermore, vessel wall pulsation in cerebral arteries has been suggested to be the primary driving force of the recently proposed glymphatic fluid flow in the brain (11–13). Thus a diminished VC may also result in other pathology, such as Alzheimer’s disease, in the brain.

Vascular stiffness of arterial vessels outside the cranium has been studied extensively. For example, tonometry and ultrasound-based methods have been applied to measure pressure or flow waves along the arteries and, by determining the time delay of these waves between two artery locations, e.g. between common carotid and femoral arteries, one can obtain the velocity of the wave propagation, known as the pulse wave velocity (PWV), which provides an indication of vascular stiffness (7,8). Ultrasound has also been used to measure vessel diameter and its change during systole and diastole of a cardiac cycle, and when comparing the vessel volume change to that of pressure, an index of VC can be computed (14). While these techniques can be applied to surface vessels, they cannot be used as effectively in intracranial vessel evaluation because of the limited ability to assess deep cerebral tissues under the overlying skull. MRI has been used to assess PWV in limited spatial locations. For example, PWV in the aortic arch can be quantified by using phase-contrast (PC) MRI, in which the temporal delay between flow waveforms in the ascending and descending aorta was determined as well as the distance along the arch (15,16). However, the above methods provide a single value of vascular stiffness, with no spatial information. More recently, Arterial-Spin-Labeling (ASL) based MRI methods have been proposed to assess differences in arterial cerebral blood volume (CBV) between systole and diastole (1,2), from which a measure of VC was obtained. While feasible, these methods require a multi-delay sampling scheme in combination with kinetic modeling to obtain a single-cardiac-phase CBV measure. As a result, the acquisition time is relatively long and, within reasonable scan duration, only single-slice 2D imaging is used (2) or spatial resolution needs to be substantially reduced (1). Therefore, a high-resolution 3D method for VC mapping is desirable.

The aim of this study is to develop an MRI technique that allows quantitative mapping of VC in major intracranial arteries. This technique, dubbed VaCom-PCASL, employs PCASL labeling (17), stack-of-stars k-space acquisition (18), and golden-angle radial sparse parallel (GRASP) reconstruction (19,20) to measure arterial vessel volume in a spatially and cardiac-phase resolved manner. This volumetric measure, when combined with the pulse pressure value, then yields an estimation of VC. Three studies were performed. Study 1 demonstrated the feasibility of the VaCom-PCASL technique and conducted sequence and reconstruction optimizations. In Study 2, intracranial VC measured with VaCom-PCASL was compared with aortic PWV measured with PC MRI. Finally, Study 3 investigated the sensitivity of the technique in detecting age effects on intracranial VC.

METHODS

Pulse sequence

The general concept of the proposed sequence is to use PCASL labeling of sufficient length to allow the labeled spins to completely fill the major arterial tree, at which time a labeled (or control) image is acquired. The subtraction of labeled and control images then yields a CBV-weighted image, with minimal CBF contributions as no post-labeling delay is used and limited labeled spins would have reached the tissue. The image intensity is then investigated as a function of cardiac phase, from which fractional changes in CBV during the cardiac cycle are obtained. Finally, based on the relationship between volume and pressure alternation, which is approximated by the brachial pulse pressure, VC is computed.

The pulse sequence of VaCom-PCASL is illustrated in Figure 1a. Following a pre-saturation to suppress tissue signals and “reset” magnetizations of vascular spins from the previous TR, a pseudo-continuous labeling module was employed to label the arterial blood. Immediately following the labeling, with a minimal post-labeling delay of 10ms during which a fat saturation pulse as well as dephasing gradient was played out, data were collected with a 3D radial acquisition method. More specifically, we used a segmented 3D radial turbo field echo sequence (21), in which the k-space was sampled using a stack-of-stars scheme where Cartesian sampling was used along the k_z direction and golden-angle radial sampling was used in the k_x-k_y plane (Figure 1a). Within each TR, 40 k lines were acquired. They were parallel to each other and arranged along the k_z direction with the same k_x-k_y coordinates. The acquisition of these 40 k lines followed a low-to-high order according to their k_z values. The labeled and control data were acquired in interleaved TRs.

Figure 1.

VaCom-PCASL pulse sequence and reconstruction methods. (a) Pulse sequence diagram and the order of k lines acquired in one TR. A TFE acquisition was used and the k lines were collected in a low-high order, as indicated by the color in the sequence and k-space diagram. (b) k-space data were retrospectively grouped into several volumes according to their cardiac phase and label/control condition. (c) Reconstruction of the k space data into cardiac-phase-resolved control and label images, the subtraction of which yielded the VaCom-PCASL images.

To fill the entire k space, across TRs, the k lines are rotated in the k_x-k_y plane by the golden angle (137.5°). With this scheme, the azimuthal sampling is approximately uniform regardless the total number of TRs. To obtain CBV-weighted image as a function of cardiac phase, a pulse oximeter (Philips Healthcare, Best, Netherlands) was attached to the index finger of the subject and the timing of cardiac pulse was recorded with reference to the pulse sequence timing (specifically, the time of the first excitation RF pulse in each TR). Then, based on the cardiac phase of each segment (also referred to as spokes), the entire k-space data can be retrospectively grouped into a number of time frames. This was done separately for control and label conditions, as illustrated in Figure 1b. The data sorting was performed such that the number of spokes in each time frame was the same. Note that, to ensure that the scan duration did not become prohibitive, we typically acquire 221 spokes for each condition. When dividing them into approximately 10 cardiac phases (see below for optimization of the number of phases), each frame has only 20–30 spokes. Thus the k space in each frame is highly under-sampled. Therefore, advanced image reconstruction schemes exploiting sparsity along the k-t space as well as the control-label conditions were employed.

Image reconstruction

A GRASP reconstruction algorithm (19,20) was used to reconstruct VaCom-PCASL images. Briefly, the reconstruction was performed by solving $x=\mathit{\text{argmin}}\left\{{\Vert F\cdot S\cdot x-k\Vert }_2^2+{\lambda }_{\mathit{\text{cardiac}}}{\Vert D_{\text{cardiac}}\cdot x\Vert }_1+{\lambda }_{\mathit{\text{control}}/\mathit{\text{label}}}\Vert D_{\text{control}/\text{label}}\cdot x{\Vert }_1\right\}$. Here F is a non-uniform fast Fourier transfer (NUFFT) operator, S is the estimated multi-coil sensitivity maps, x is the 5D image series we desire, with three spatial dimensions, one cardiac-phase dimension, and one label/control dimension. k is the multi-coil radial k-space data mentioned above, D_cardiac and D_{control/label} are the difference operators that compute image differences between consecutive images along the cardiac phase (e.g. cardiac phase 2 – cardiac phase 1; cardiac phase 3 – cardiac phase 2; and so on) and control/label (i.e. control – label for each cardiac phase) dimensions, respectively. λ_cardiac and λ_{control/label} are corresponding regularization weights that control the tradeoff between data fidelity and sparsity constraints. The multi-coil sensitivity maps were obtained by dividing single-coil image by the root of sum of squared image of all coils, when using data from all frames. The GRASP reconstruction yields a series of control/label images at different cardiac phases (Figure 1c), from which a series of cardiac-phase-resolved VaCom-PCASL images were obtained.

Quantification of vascular compliance

The VaCom-PCASL signal (i.e. control-label) can be written as:

$$S=2\cdot M_0\cdot \alpha \cdot CBV\cdot e^{-\delta /T_1}$$ [1]

where M₀ is the equilibrium magnetization of the blood, T1 is the blood T1, α is the labeling efficiency, and δ is the bolus arrival time. The signal intensity was studied as a function of cardiac phase and fitted to a second-order Fourier function (with 5 coefficients corresponding to sin(θ), cos(θ), sin(2θ), cos(2θ), and a constant term, respectively). The difference (ΔS) between maximum and minimum along the fitted curve corresponds to vessel pulsation (i.e. ΔCBV) during systolic and diastolic phases. The mean signal of the curve ($\overline{S}$) is proportional to baseline CBV. Therefore, voxel-by-voxel VC can be calculated accordingly by:

$$VC=\frac{\Delta S/\overline{S}}{\Delta P}\cdot 100\%$$ [2]

with the units of % volume change per mmHg of blood pressure change. Here ΔP is the pulse pressure (i.e. systolic – diastolic blood pressure) measured at the brachial artery, which provides a surrogate of pulse pressure in the intracranial vessels.

Simulations to investigate the confounding effect of bolus arrival time (δ) on VC estimation

In the VC quantification formula described in Eq. [1], it was assumed that the fluctuation of S during the cardiac cycle was solely attributed to CBV. In reality, however, the flow velocity in the major arteries of the brain also varies within the cardiac cycle, thus δ may also change as a function of the cardiac phase. To examine this effect, simulations were conducted in which cardiac-phase dependent flow velocity in large vessels was assumed based on the report of Wu et al. (22). Then, we studied a range of travel distances (10–15cm) that the spin has to travel from the PCASL labeling location to the vessel of interest. For each cardiac-phase in the VaCom-PCASL data, we calculated δ using the area-under-curve (preceding the acquisition time) of the flow-velocity time-course, from which the variation in δ between diastolic and systolic phase was computed. The impact of δ on VC estimation was then estimated using Eq. [1].

MRI experiments

MRI experiments were carried out using a 3T MRI scanner (Philips Healthcare, Best, Netherlands). The RF transmission used the body coil, and the signal reception used a 32-channel head coil. Foam padding was used to stabilize the head and minimize motion. The protocol was approved by the Institutional Review Board of Johns Hopkins University. All subjects gave informed written consent before participating in the study. Three sub-studies were conducted.

Study 1: Feasibility and optimization of VaCom-PCASL MRI

Eight healthy volunteers (43.3±20.8 years, 4 males) were scanned in this study. The imaging parameters were: FOV=180×180×80 mm³, voxel size=2×2×2 mm³, 40 partitions along k_z, 221 spokes in k_x-k_y plane, TFE readout, TFE factor=40, 442 TRs including both control and label conditions, time between TFE readouts=4.7 ms, TE =2.1 ms, flip angle=9°, duration of all TFE readouts=189 ms. In three initial feasibility-testing subjects, a labeling duration of 800 ms was used. In the later five subjects, four different labeling durations, 400, 800, 1200, and 1600 ms, were used to investigate the effect of the labeling duration on the results. After the MRI session, brachial pulse pressure, ΔP, was measured outside the scanner room by cuff blood pressure device (Omron Corporation, Kyoto, Japan). Two measurements were performed and the values were averaged.

We first sought to determine the optimal number of cardiac phases, N_cardiac, and weighting parameters, λ_{control/label} and λ_cardiac, and in the GRASP reconstruction. For the number of cardiac phases, one aims to achieve a balance between sufficient temporal resolution within the cardiac cycle and enough k-space data within each phase to minimize streaking artifacts. Five N_cardiac values, 7, 10, 13, 17, and 22, were tested. For the regularization weights, λ_{control/label} and λ_cardiac, five values, 0.0N₀, 0.01N₀, 0.05N₀, 0.1N₀, and 0.3N₀, were tested where N₀ was the maximal magnitude value of the NUFFT images. Using the data with 800 ms labeling duration (N=8), GRASP reconstruction was performed using in-house MATLAB (MathWorks, Natick, MA, USA) scripts based on the work of Feng et al. (https://www.cai2r.net/grasp) (19,20). All reconstructions were performed on a Linux Cluster (AMD Opteron Processor 6320, 128 GB RAM). For each parameter set, a contrast-to-noise ratio (CNR) was calculated in which the contrast was the signal difference between systolic and diastolic phases in an ROI containing the arteries (threshold at 10% of maximum VaCom-PCASL signal intensity), and the noise was the temporal (across cardiac phases) standard deviation of the signal in a tissue region without visible vessels. The parameter set that yielded the highest CNR was identified as the optimal reconstruction parameter.

Next, we examined the impact of labeling duration. An optimal labeling duration is such that it is long enough to allow labeled water spins to completely fill the artery tree while minimizing the scan time. VaCom-PCASL data collected with labeling durations (LD) of 400, 800, 1200, and 1600 ms were compared for signal intensities, using signals in the LD=1600 ms scan as a reference. The scan duration for these sequences was 5 min 5 sec, 10 min 40 sec, 14 min 44 sec, and 19 min 30 sec, respectively.

Statistical analyses were performed using MATLAB (The MathWorks, Inc., Natick, MA). In all analyses, a p-value of 0.05 or less was considered significant.

Study 2: Comparison of VaCom-PCASL with aortic vascular stiffness assessed with pulse wave velocity (PWV)

Aortic PWV measured by phase-contrast (PC) MRI has been used as an index of arterial stiffening in the literature, in which higher PWV indicates stiffer vessels (23,24). In order to corroborate the VC determined by VaCom-PCASL with an independent technique, we compared our measure of intracranial VC with aortic PWV in eight subjects (43.9±15.8 years, 4 males). The VaCom-PCASL sequence used parameters optimized in Study 1. The PWV scan followed protocol used by King et al. (23). Briefly, we first performed a breath-hold, oblique, double inversion-recovery turbo spin-echo sequence to visualize the ascending and descending segments of the aorta (Figure 2a). Then, a cardiac-gated, breath-hold PC sequence was performed in the transverse plane at the level of the pulmonary artery, which allows the imaging of the ascending and descending aorta simultaneously (Figure 2a). The imaging parameters were as follows: velocity-encoding=200 cm/s, FOV=300×250 mm², voxel size=2.5×2.5 mm², TR=4.2 ms, TE=2.7 ms, and 40 phases per cardiac cycle. Regions-of-interest (ROIs) of ascending and descending aorta were manually drawn (Figure 2b) on complex difference images averaged across all heart phases. Blood flux value at each vessel and each phase was obtained by applying the ROIs onto velocity maps (Figure 2c), yield an ascending and a descending flux time curve (Figure 2d). The time it takes for the pulse wave to travel from ascending to descending aorta was calculated as the relative shift between the two curves, determined by time-shifting and finding the best cross-correlation between the two curves. To determine the distance the blood has travelled, using the ImageJ software (National Institutes of Health, Bethesda, MD), the aortic arch distance between ascending and descending aortic PC imaging site was determined by drawing a freehand line through the center of the aorta parallel to the aortic walls (Figure 2a). PWV was calculated by dividing the aortic arch distance by the time shift (25). VC obtained from VaCom-PCASL and PWV from PC MRI were compared using Pearson correlation.

Figure 2.

Measurement of aortic PWV using PC MRI. (a) Sagittal view of aorta showing the positioning of PC imaging slice (orange line) and the distance between ascending and descending locations (white dash curve). (b-c) Magnitude and phase images of PC MRI showing ascending (red) and descending (blue) aorta. (d) Normalized (to peak) flux time courses in ascending (red) and descending (blue) aorta. N=8. Transit time was estimated by shifting flux curve of the descending aorta to achieve maximum cross-correlation with flux curve in the ascending aorta. PWV was calculated by dividing the travel distance by the transit time.

Study 3: Sensitivity of VaCom-PCASL to aging

Aging is major risk factor for many cerebrovascular diseases and is associated with increased arterial stiffness (26). We therefore tested the sensitivity of VaCom-PCASL by comparing intracranial vascular compliance between young and older subjects. A total of seven younger (age 30.7±6.9 years) and seven older (age 61.7±7.4 years) healthy volunteers were included. Of these, 5 (3 young, 2 older) were from Study 1, 8 (4 young, 4 older) were from Study 2, and 1 (older) was scanned separately. The optimal VaCom-PCASL protocol as determined in Study 1 was used.

Quantitative comparison of VC values in younger and older participants was performed on regions-of-interest (ROIs) corresponding to internal carotid arteries (ICA), anterior cerebral arteries (ACA), middle cerebral arteries (MCA), and posterior cerebral arteries (PCA). To obtain the ROIs, an arterial mask of the whole brain was first generated by thresholding the averaged VaCom-PCASL image at 10% of its maximum signal intensity. Next, voxels corresponding to the specific artery were manually delineated from the arterial mask. By applying the ROI onto the VaCom-PCASL image series, a signal intensity time curve was obtained and the corresponding VC value was quantified based on the method described above. VC values were compared between the younger and older groups using two sample t-tests. Correlation between VC and mean blood pressure was studied.

RESULTS

Study 1: Feasibility and optimization of VaCom-PCASL MRI

Figure 3a shows representative control and label images from the VaCom-PCASL sequence (at a systolic phase). It can be seen that signals in the tissue regions are largely the same between control and label, but the large vessel signals were bright in control images and dark in label images. MIPs of the VaCom-PCASL (i.e. difference) images are also displayed in Figure 3a.

Figure 3.

Representative results of VaCom-PCASL. (a) Reconstructed control, label, and VaCom-PCASL MIP images during a systolic phase. (b) VaCom-PCASL signal intensity of an MCA segment (red region in the inset) as a function of cardiac phase and its corresponding fitted curve (2nd-order Fourier series). The synchronized peripheral pulse (PPU) signal is also shown. (c) Sagittal, coronal, and axial MIP of phase-averaged VaCom-PCASL image ($\overline{S}$), VaCom-PCASL fluctuation (ΔS), and vascular compliance (VC) map, respectively.

Figure 3b shows representative VaCom-PCASL signal fluctuations as a function of cardiac phase in an ROI containing the MCA, exhibiting a clear pulsatile modulation. The corresponding peripheral pulse (PPU) signal is also shown. Figure 3c displays the cardiac-phase-averaged VaCom-PCASL images ($\overline{S}$), signal fluctuation (ΔS) map, and the corresponding VC map. Note that both averaged VaCom-PCASL and fluctuation maps show higher values in larger vessels (red in color), but their effects are canceled out in the VC map (far right in Figure 3c). Supporting information Video S1 exhibits the mean VaCom-PCASL image, VaCom-PCASL images as a function of cardiac phase, and the corresponding 3D VC map.

The performance dependence of the technique on reconstruction parameters, specifically the number of cardiac phases N_cardiac (see Supporting Information Figure S1 for an example), and the two regularization weights, λ_{control/label} and λ_cardiac, was examined. As shown in Figure 4, N_cardiac=10, λ_{label/control}=0.05N₀, and λ_cardiac=0.05N₀ provided the highest CNR. These reconstruction parameters were used in later studies.

Figure 4.

Dependence of contrast-to-noise ratio (CNR) on reconstruction parameters. CNR was computed as arterial VaCom-PCASL signal fluctuation across cardiac phase divided by temporal standard deviation of tissue signal. (a) Number of cardiac phases, N_cardiac, (b) Control/label weighting factor, λ_{label/control}, (c) Cardiac phase weighting factor, λ_cardiac. N=8.

Since the number of spokes in each time frame was forced to be the same in our reconstruction algorithm, we further examined the degree to which the actual cardiac phases deviated from the assigned phases. Figure 5 displays the mean phase of spokes in an image for every cardiac phase and every subject, relative to the assigned phase. It can be seen that the deviations are generally small and are distributed around zero (p>0.05 for all phases).

Figure 5.

Difference between mean cardiac phase of spokes in each frame and its assigned phase, for control and label images, respectively. N=8.

The dependence of the technique on labeling duration was also investigated. We separately analyzed ROIs containing larger vessels (arteries within or near the Circle of Willis) and smaller more distal arteries (pink and green, respectively, in Figure 6a). Figure 6b shows the dependence of $\Delta S/\overline{S}$ on labeling duration. It can be seen that $\Delta S/\overline{S}$ was overestimated when using a short labeling duration of 400 ms, in both larger and smaller arteries (p<0.005). This can be attributed to “inflow”, in addition to “volume”, pulsation between systolic and diastolic phases. That is, part of the signal variation is due to the arrival of the labeled blood in some but not in other cardiac phases. The overestimation is more pronounced in smaller (i.e., more distal) arteries because of their longer bolus arrival time and thus greater inflow-related signal variations across cardiac cycle. Note that we only want to measure the effect of “volume” pulsation in VC, and thus the inflow effects are not desired. $\Delta S/\overline{S}$ plateaued at a labeling duration of 800 ms, as there was not a significant difference between 800 ms and longer labeling durations (p>0.2). However, it should be noted that older individuals are expected to have a longer arterial bolus arrival time. Therefore, for the aging study (i.e. Study 3), we used a labeling duration of 1200 ms to account for potentially longer arterial bolus arrival time in this population (27).

Figure 6.

Effect of labeling duration on VaCom-PCASL. (a) ROIs of larger and distal smaller arteries used in the quantitative analysis. (b) Fractional fluctuations in VaCom-PCASL signal ($\Delta S/\overline{S}$) in larger and smaller arteries as a function of labeling duration. N=5.

Simulation results examining the effect of δ are shown in Figure 7. Figure 7a shows the difference in δ between diastolic and systolic phase as a function of travel distance. Figure 7b shows the estimated $\Delta S/\overline{S}$ signal when assuming a true $\Delta CBV/\overline{CBV}$ of 10.4% (based on our experimental data in Figure 3b). As can be seen, within the travel distance examined, the maximum estimation bias is 1.0% (about 10% of the CBV effect).

Figure 7.

Simulations of the effect of bolus arrival time (δ) on VC estimation. (a) Difference in δ between diastolic and systolic phase as a function of travel distance. (b) Estimated $\Delta S/\overline{S}$ signal when assuming a true $\Delta CBV/\overline{CBV}$ of 10.4%.

Study 2: Relationship between intracranial vascular compliance and aortic pulse wave velocity

Figure 2d shows group-averaged phase-contrast flux results in both ascending and descending aorta. Figure 8 shows the scatter plot of VC at ICA and aortic arch PWV. The VC of ICA exhibited a significant negative correlation with aortic arch PWV (r=−0.73, p=0.038). No correlations were found between $\Delta S/\overline{S}$ and aortic arch PWV (p=0.85), or between pulse pressure and PWV (p=0.44).

Figure 8.

Scatter plot between VC of internal carotid artery and aortic PWV across subjects. N=8.

Study 3: Aging effect on intracranial vascular compliance

Table 1 summarizes the blood pressure, VaCom-PCASL signal, and VC results for the younger and older groups. Representative VC maps from an older (age 75 years, female) and a younger (age 30 years, male) participant are shown in Figure 9a. It can be seen that vascular compliance in the older subject appears to be lower throughout the arterial tree. Figure 9b shows the ROI results. Vascular compliance in the older group was generally lower when compared to the younger group, with ICA, ACA, and MCA results showing a significant difference. There was an inverse correlation between VC and mean blood pressure (r=−0.59, p=0.0251 for ICA; r=−0.89, p<0.0001 for ACA; r=−0.72, p=0.004 for MCA; r=−0.74, p=0.0027 for PCA). It was also noted that, within the same subject, VC in ICA was lower than that of its downstream branches, i.e. ACA, MCA, and PCA (p<0.01). This is consistent with the notion that smaller arterial vessels have a greater dilatory capacity.

Table 1.

Demographic information of Study 3 participants and VaCom-PCASL results (mean ± standard error).

		Age groups
		Young	Old	p-value
No. of participants		7 (4 males)	7 (2 males)
Age (year)		30.7 ± 2.6	61.7 ± 2.8	<0.001
Systolic BP (mmHg)		105.4 ± 2.6	132.6 ± 4.5	<0.001
Diastolic BP (mmHg)		77.8 ± 2.5	79.6 ± 4.1	0.72
Mean BP (mmHg)		87.0 ± 2.5	97.2 ± 3.6	0.039
Pulse pressure (mmHg)		27.6 ± 1.6	53.0 ± 4.7	<0.001
$\Delta S/\overline{S}$ (%)	ICA	5.91 ± 0.66	8.48 ± 1.21	0.087
	ACA	9.98 ± 1.41	11.99 ± 2.34	0.48
	MCA	9.33 ± 1.32	10.49 ± 1.28	0.54
	PCA	10.55 ± 1.43	14.76 ± 2.78	0.21
VC (%/mmHg)	ICA	0.214 ± 0.020	0.156 ± 0.010	0.025
	ACA	0.359 ± 0.040	0.228 ± 0.042	0.044
	MCA	0.335 ± 0.036	0.203 ± 0.028	0.013
	PCA	0.378 ± 0.038	0.283 ± 0.053	0.18

Figure 9.

Age dependence of intracranial VC. (a) Representative VC maps from an elderly (75 years, female) and a young (30 years, male) subject. (b) Comparison of VC in ROIs corresponding to ICA, ACA, MCA, and PCA between young (N=7) and elderly (N=7) participants. Error bars indicate the standard error. *p<0.05. An illustration of the ROIs is shown in (c).

DISCUSSION

In this study, a new MRI technique, VaCom-PCASL, was presented for 3D mapping of intracranial vascular compliance without the use of exogenous contrast agent. The technique utilizes cardiac pulsation as an intrinsic pressure stimulus and measures changes in blood vessel caliber within a cardiac cycle using ASL-based radial acquisition and compressed sensing reconstruction. We demonstrated that this technique was able to measure vascular compliance along the major branches of intracranial arterial tree. We compared the VaCom-PCASL technique with an established aortic stiffness measurement and found a significant correlation between intracranial VC and aortic PWV. Using the proposed technique, age differences in VC were studied, and older individuals revealed a significantly lower vascular compliance when compared with younger participants.

Clinical considerations

Vascular compliance, which reflects the stiffness of an arterial vessel wall, is an important property of an artery in the presence of pulsatile blood flow that originates from the heart, and is crucial in attenuating flow pulsatility in large vessels into continuous flow in the distal capillary bed, thus protecting blood-brain barrier (BBB) and brain parenchyma from mechanical injury. Arterial stiffness has been proposed to be an important biomarker in several pathological conditions, including hypertension (28), diabetes (29), cardiovascular disease (30), and a variety of cerebrovascular diseases (e.g. stroke (5), small-vessel disease (31)). At present, arterial stiffness along aortic, common carotid, common femoral, or brachial arteries, can be assessed in clinical settings by using primarily an ultrasound echo-tracking method. However, the assessment of intracranial vascular stiffness using ultrasound methods remains challenging due to the presence of the barrier of the overlying skull. Therefore, the development of non-invasive, non-contrast VC techniques such as VaCom-PCASL can potentially fill this gap and may have clinical applications in cerebrovascular diseases.

Additionally, arterial pulsatility, $\Delta S/\overline{S}$, has been postulated to be the primary driving force of the recently discovered glymphatic system of the brain, in which arterial wall pulsation results in directional fluid movement in cerebrospinal fluid (CSF) and the interstitial space (11–13). The glymphatic system is thought to play a critical role in the “waste clearance” of the brain and has been suggested to be related to several neurodegenerative diseases. Therefore, a non-invasive method to quantify the driving force of the glymphatic system may be clinically useful.

Comparison with other techniques

Several previous methods have been proposed to measure vascular compliance. Warnert et al. measured the cross-sectional area of MCA in diastole and systole using cardiac-triggered T2-weighted sequence at 7 Tesla (32), and reported an absence of correlation between MCA distension and the flow pulsatility index as determined from phase-contrast MRI. Recently, two ASL-based MRI techniques have been developed to estimate the intracranial CBV change between systole and diastole. Yan et al. (2,33) demonstrated that, by tracking ASL signal time course with multi-phase bSSFP after pulsed labeling, one can quantify CBV. When CBV during systole and diastole are separately obtained using cardiac triggering, VC can be computed. One limitation of this technique is that it is at present only implemented for single-slice acquisition, which resulted in reduced spatial coverage. Its temporal resolution was also limited in that only two phases (systole and diastole) were acquired and thus required an additional time-resolved PC to determine the trigger delay for these specific cardiac phases. Another ASL-based method (1) estimated CBV by fitting multi-delay pulsed ASL data to a kinetic model, and CBV of diastole (systole) was obtained after binning the ASL data retrospectively to diastolic (systolic) phase. This method had limited spatial (3×3×7 mm³) resolutions. The VaCom-PCASL technique presented in this study is based on single-time measurement of CBV and does not rely on kinetic modeling and fitting. By taking advantage of k-space undersampling using radial stack-of-star acquisition as well as compressed sensing reconstruction, relatively high spatial (2×2×2 mm³) and temporal (10 cardiac phases) resolution and coverage (180×180×80 mm³) were obtained.

Technical considerations

Due to the intrinsic incoherent aliasing, radial k-space filling schemes are ideally suited for compressed sensing reconstruction to accelerate data acquisition in dynamic MRI. Moreover, the repeated sampling of k-space center in radial acquisition improves motion robustness. In the present study, a stack-of-stars sampling scheme was used which employs undersampling (and thereby compressed sensing) in the k_x-k_y dimensions but uses full sampling along the k_z dimension. The reason that we did not use undersampling in all three dimensions (34) (which is known as the 3D radial koosh-ball scheme) is that, since we are using TFE to accelerate the acquisition, a stack-of-stars scheme can be easily made compatible with TFE by a low-high frequency order. The low-high order allows the center of the k-space to have the maximal signal within intracranial arteries. In our sequence, the TFE scheme allowed the acquisition of all k_z-lines within one TR.

The golden-angle acquisition scheme has previously been applied for accelerated imaging of the heart, brain, and other organs (35–40). It allows approximately uniform coverage of the k-space for any arbitrary number of consecutive spokes. In the proposed VaCom-PCASL technique, the acquired spokes were retrospectively grouped according to its cardiac phase, which makes the spoke arrangement in each cardiac phase not strictly uniform. However, since we used a GRASP algorithm for dynamic image reconstruction, information in spokes of adjacent cardiac phases and in both control and labeled data has also been used.

Labeling duration is a critical sequence parameter for VaCom-PCASL, which should be chosen to be long enough for the labeled blood water to fill the intracranial artery tree but not make the scan duration prohibitively long. If the labeling duration is too short, downstream smaller arteries will not be visible, and furthermore overestimation in VC due to inflow pulsatility effects will occur (as shown in Study 1). That is, the bolus may have arrived during the systole phase while not the case during diastole phase. On the other hand, for labeling duration of 1600 ms, the scan time is almost 20 minutes, making it a challenge for application in clinical populations.

Compared with standard time-of-flight MRA sequence, a relatively large voxel size (2×2×2 mm³) was used in the proposed VaCom-PCASL technique. This was to ensure that the imaging voxel always contains some non-blood tissue components. If a voxel is 100% blood, e.g. a small voxel located exclusively in the lumen of an artery, the VaCom-PCASL signal intensity would not show a change between systolic and diastolic phases even if the artery dilates.

The mechanism of the VaCom-PCASL signal is primarily attributed to cardiac-phase dependent changes in CBV. However, as indicated in Eq. [1], alternations in bolus arrival time could also contribute to the signal. Our simulations suggested that this effect is relatively small and its contribution is 0.5%-9.5% of the CBV effect, depending on the travel distance.

This technique does not require cardiac triggering with ECG. Instead, a finger pulse oximeter was used to record the pulse timing with reference to data acquisition. The simple setup of the technique makes it feasible for convenient clinical applications. Additionally, the proposed technique has the flexibility of making sure all k-space data is collected regardless their cardiac phase without need to discard any data, unlike for instance cardiac-gated phase-contrast sequence.

Limitations

One limitation of the VaCom-PCASL technique is that we used a single pulse pressure measure (i.e. at the brachial artery) as a surrogate for pulse pressure of all intracranial segments, whereas it is expected that the pulse pressure will decrease in distal arteries. Furthermore, the drop in pulse pressure may be different in different arterial branches, in that a branch with stiffer vessels may have a lesser decrease, resulting in an over-estimation in VC. This may reduce the sensitivity of the method in detecting vessel stiffness abnormalities. Unfortunately, to our knowledge, there is no available method to non-invasively measure pulse pressure in intracranial vessels. This is a limitation of virtually all brain vascular compliance measurement methods in the field. Additionally, in our calculation, we have not accounted for cardiac-phase dependent variations in intracranial pressure. While these variations are expected to be much smaller than that of intra-arterial pressure, they would nonetheless result in an under-estimation in VC. Another limitation is that the assigned cardiac phases for reconstruction were slightly different from the actual values (Figure 5), due to our strategy to fix spoke number in each cardiac phase. Similarly, the actual phases of the control and labeled images may be different, which will affect the estimation of VaCom-PCASL signal. An alternative reconstruction strategy is to force the cardiac phase bins to be fixed, but allow the number of spokes in each phase to vary. This strategy would allow the assigned cardiac phase to be more accurate, but at the expense of image quality (when the number of spokes in a particular cardiac phase is excessively low). However, it should be noted that, as the total number of spokes collected increases, these two methods will yield converging results. Finally, in the TFE acquisition, a fixed flip angle was used and the saturation effects of early TFE excitation pulses on later TFE data were not considered. Future studies should consider the varying-flip-angle scheme in the TFE acquisition.

CONCLUSION

We demonstrated that pulse-pressure-induced changes in arterial vessel caliber can be quantified with a VaCom-PCASL technique using PCASL labeling, radial acquisition, and compressed sensing reconstruction. This method provides a non-invasive assessment of arterial wall pulsatility and vascular compliance in the brain, and may have potential utility in studies of the glymphatic system, as well as cardiovascular and cerebrovascular diseases.

Supplementary Material

figS1

Supporting Information Figure S1. Representative images of VaComp-PCASL reconstructed using different numbers of cardiac phases. The images shown are from a systolic phase of a subject. The number of spokes corresponding to each phase number is also displayed.

videoS1

Supporting Information Video S1. Three-dimensional display of VaCom-PCASL images. (a) Cardiac-phase-averaged VaCom-PCASL image. (b) Cardiac-phase related variations in VaCom-PCASL image. To allow easy visualization of temporal fluctuations in the presence of spatial signal heterogeneities, the fluctuation amplitude of each voxel was augmented by 5 times. (c) VC map.