Noncontrast assessment of blood–brain barrier permeability to water: Shorter acquisition, test–retest reproducibility, and comparison with contrast-based method

Abstract

Purpose:

Assessment of the blood–brain barrier (BBB) permeability without the need for contrast agent is desirable, and the ability to measure the permeability to small molecules such as water may further increase the sensitivity in detecting diseases. This study proposed a time-efficient, noncontrast method to measure BBB permeability to water, evaluated its test–retest reproducibility, and compared it with a contrast agent–based method.

Methods:

A single-delay water extraction with phase-contrast arterial spin tagging (WEPCAST) method was devised in which spatial profile of the signal along the superior sagittal sinus was used to estimate bolus arrival time, and the WEPCAST signal at the corresponding location was used to compute water extraction fraction, which was combined with global cerebral blood flow to estimate BBB permeability surface area product to water. The reliability of WEPCAST sequence was examined in terms of intrasession, intersession, and inter-vendor (Philips [Ingenia, Best, the Netherlands] and Siemens [Prisma, Erlangen, Germany]) reproducibility. Finally, we compared this new technique to a contrast agent–based method.

Results:

Single-delay WEPCAST reduced the scan duration from approximately 20 min to 5 min. Extract fraction values estimated from single-delay WEPCAST showed good consistency with the multi-delay method (R = 0.82, P = .004). Group-averaged permeability surface area product values were found to be 137.5 ± 9.3 mL/100 g/min. Intrasession, intersession, and inter-vendor coefficient of variation of the permeability surface area product values were 6.6 ± 4.5%, 6.9 ± 3.7%, and 8.9 ± 3.0%, respectively. Finally, permeability surface area product obtained from WEPCAST MRI showed a significant correlation with that from the contrast-based method (R = .73, P = .02).

Conclusion:

Single-delay WEPCAST MRI can measure BBB permeability to water within 5 min with an intrasession, intersession, and inter-vendor test–retest reproducibility of 6% to 9%. This method may provide a useful marker of BBB breakdown in clinical studies.

1 |. INTRODUCTION

Blood–brain barrier (BBB) is a selective and tight barrier formed by endothelial cells, astrocytes, and pericytes.¹ It regulates the Blood–brain exchange of substances and maintains a constant microenvironment that is crucial for normal function of the nervous system. Alterations in BBB permeability have been associated with many disorders, such as Alzheimer disease,² multiple sclerosis,³ stroke,⁴ and brain tumor.⁵

Contrast agent-based MRI method has been used to evaluate BBB permeability to gadolinium (Gd) chelates — qualitatively by examining postcontrast T₁ enhancement or quantitatively with dynamic contrast-enhanced MRI.⁶ Although this method can be used to measure major BBB breakdowns, its sensitivity to small disruption of BBB at the early stage of disease is relatively low,^7–9 and its acquisition time is usually long. Additionally, injection of Gd contrast agent can be complicated by difficulty obtaining intravenous access as well as the possibility of contrast extravasation, allergic reaction to contrast, nephrogenic systemic fibrosis, and potential negative impact of Gd deposition in the brain.^10,11

Recently, there has been an increasing interest in examining the BBB’s permeability to water, an endogenous tracer with a smaller molecular size.^12,13 It is known that water is not freely diffusible across the BBB in healthy brain (eg, unlike [¹¹C]Butanol),^14–18 and thus quantification of water permeability may provide information about subtle change in BBB integrity. Several MRI methods have been proposed to measure BBB water permeability by separating intravascular and extravascular water signal based on their differences in spatial location,¹⁹ T₁,^17,20,21 T_2,^22–25 or diffusion properties.^{13,18,26–32} Other methods using contrast agent also exploited the sensitivity to water permeability.^33–36 However, a limitation of the previous methods is that the measurements often involved multiple scans with a long acquisition time and thus may present an obstacle for clinical applications.

Water extraction with phase-contrast arterial spin tagging (WEPCAST) MRI is a recently proposed noncontrast technique that provides a measurement of global water extraction fraction (E) and BBB permeability surface area product (PS) by selectively measuring arterial spin labeling (ASL)^37–39 signals in venous vessels (thereby evaluating the rate of water exchange across BBB).¹⁹ The approach used in an earlier proof-of-principle study¹⁹ required the acquisition of signals at multiple post-labeling delay times (PLD), which resulted in a total scan time of more than 20 min. The goal of the present study is therefore to develop a clinically applicable, single-delay method of WEPCAST MRI with a target scan time of approximately 5 min. We conducted 3 substudies to develop and evaluate this method. In study I, we developed theoretical, experimental, and analysis procedures for single-delay WEPCAST and compared the data to multi-delay results. In study II, the reliability of the single-delay WEPCAST technique was characterized by evaluating its intrasession and intersession reproducibility on the same scanner as well as reproducibility across scanners of different vendors. Finally, in study III we compared this noncontrast BBB technique of water permeability to a contrast agent–based method.

2 |. METHODS

All procedures in this study have been approved by the institutional review board of the Johns Hopkins University School of Medicine, Baltimore, Maryland. Written informed consent was obtained before the participants were enrolled in the study. A total of 29 healthy participants were studied. The experiments were performed on a 3 Tesla Philips system (Ingenia, Best, the Netherlands), except for the cross-vendor study in which a 3 Tesla Siemens system (Prisma, Erlangen, Germany) was additionally used.

2.1 |. Study I: Development of Single-delay WEPCAST MRI

2.1.1 |. Pulse sequence and theory

The pulse sequence of WEPCAST MRI is displayed in Figure 1A. This sequence labels water in the arterial blood at the cervical spine (Figure 1B) with a pseudo-continuous ASL module. At the capillary–tissue interface, water exchanges across BBB and a fraction of the labeled water molecules are extracted into the tissue, whereas the remaining unextracted water is drained to the venous vessels (Figure 1C). With this fundamental concept, WEPCAST MRI employs a velocity-sensitive gradient to selectively measure the amount of labeled water in draining veins such as the superior sagittal sinus (SSS) (Figure 1B), which drains most cerebral tissues, yielding an estimation of global extraction fraction of water (E) (Figure 1C). Together with global cerebral blood flow (CBF), BBB permeability to water, denoted by PS, can be estimated as⁴⁰:

$$PS=-\text{ln\hspace{0.17em}}(1-E)\cdot CBF.$$ (1)

FIGURE 1.

WEPCAST MRI technique for noncontrast assessment of BBB permeability. (A) WEPCAST MRI sequence diagram. (B) Typical sagittal and coronal imaging position of WEPCAST MRI. (C) Water exchange model across BBB. E can be estimated from venous ASL signal. (D) Illustration of the spatial distribution of WEPCAST signal bolus on superior sagittal sinus for a hypothetical subject. Due to dispersion, the bolus has a higher signal in the middle (yellow) and lower signal on the edges (red). ASL, extraction fraction; BBB, blood–brain barrier; E, extraction fraction; WEPCAST, water extraction with phase-contrast arterial spin tagging

The WEPCAST signal (ie, control – label) in the vein can be written as

$$\Delta M=\Delta M_a+\Delta M_b,$$ (2)

where

$$\Delta M_a(t)=2\alpha (1-E)M_{0,\text{blood}}{e}^{-\frac{{\delta }_v}{T_{1,\text{blood}}}}{d}_a\left(t,{\delta }_v\right),$$ (3)

denotes the spins that are not extracted by tissue and drained directly to the vein (Figure 1C, blue); d_a (t, δ_v) = c(t, δ_v) ⊗ G(t);

$$\Delta M_b(t)=2\alpha \frac{f}{\lambda }E{M}_{0,\text{blood}}{e}^{-\frac{{\delta }_v}{T_{1,\text{blood}}}}{d}_b\left(t,{\delta }_v\right),$$ (4)

represents the extracted spins that are re-exchanged into the vessel (Figure 1C, yellow); and d_b(t, δ_v) = c (t, δ_v) ⊗ [r(t)m(t)] ⊗ G(t). In Equations (3) and (4), M_0,blood is the equilibrium magnetization of blood and can be experimentally determined; δ_v is the bolus arrival time for vein; f is CBF; $c\left(t,{\delta }_v\right)=\{\begin{array}{ll}1,\hfill & \hfill \\ \hfill & 0,\hfill \end{array}\begin{array}{c}\text{if\hspace{0.17em}}{\delta }_v<t<{\delta }_v+\tau \\ \text{otherwise}\end{array}$ is the arterial input function; τ is labeling duration; r(t) = e^−ft/λ is the residue function; λ is the blood–brain partition coefficient; $m(t)=e^{-t/T_{1,\mathit{\text{tissue}}}}$ represents the T₁ relaxation of the signal; and $G(t)=\frac{1}{\sqrt{2\pi {\sigma }^2}}{e}^{-t^2/2{\sigma }^2}$ denotes the Gaussian kernel due to dispersion of the labeled bolus. σ is the SD of the distribution and, based on the literature of arterial value of approximately 0.25 s,⁴¹ we assumed a venous σ of 0.5 s (ie, twice the arterial value). ⨂ denotes convolution operation, and α is the labeling efficiency. Aside from the assumed bolus dispersion kernel parameters, the only unknowns in Equations (2–4) are E and δ_v.

In the previous proof-of-principle report,¹⁹ the approach to solve for 2 unknowns in 1 equation was to perform WEPCAST experiments at several delay times (t) and then fit the signals to the model and simultaneously estimate both parameters. Here, our approach is to solve for δ_v and E sequentially using data from 1 delay time. First, rather than using the previous approach to select a spatial location on the vein and ask what the δ_v of this voxel is, we took advantage of the fact that we measured WEPCAST signal at multiple locations along the SSS (Figure 1D) and ask where the location is along the vein that has a known δ_v. Specifically, when performing a single-delay experiment with a delay time of PLD, the unextracted WEPCAST signal (ie, Equation (3)) at different locations (l) can be written as:

$$\Delta M_a(PLD+\tau ,l)=2\alpha (1-E)M_{0,\text{blood}}(l)e^{-\frac{{\delta }_v(l)}{T_{1,\text{blood}}}}{d}_a\left(PLD+\tau ,{\delta }_v(l)\right),$$ (5)

where l is the distance from a reference location that has a δ_v of $PLD+\frac{\tau }{2}$, that is, middle of the labeled bolus. By assuming a flow velocity in SSS, V, we can rewrite the Equation (5) as

$$\Delta M_a(PLD+\tau ,l)=2\alpha (1-E)M_{0,\text{blood}}(l)e^{-\frac{PLD+\frac{\tau }{2}+\frac{l}{V}}{T_{1,\text{blood}}}}{d}_a\left(PLD+\tau ,PLD+\frac{\tau }{2}+\frac{l}{V}\right).$$ (6)

We can further reformat Equation (6) into

$$\frac{\Delta M_a(PLD+\tau ,l)e^{\frac{\frac{1}{V}}{T_{1,\text{blood}}}}}{M_{0,\text{blood}}(l)}=2\alpha (1-E)e^{-\frac{PLD+\frac{\tau }{2}}{T_{1,\text{blood}}}}{d}_a\left(PLD+\tau ,PLD+\frac{\tau }{2}+\frac{l}{V}\right).$$ (7)

As such, we have transformed a function of t into a function of l (see Figure 1D for an illustration of expected signals). Plotting the function d_a (Supporting Information Figure S1), we can see that the term on the right-hand side of Equation (7) achieves its largest value when l = 0. In other words, when we identify the highest signal amplitude for the left-hand side term along the spatial profile of SSS, that location should represent l = 0 and has a δ_v of $PLD+\frac{\tau }{2}$.

Once δ_v is known, E can then be estimated using the signal amplitude (ΔM) at that location and Equation (3). Because ΔM_b is expected to be about 1 order lower than ΔM_a, its contribution is neglected in the single-delay WEPCAST method. To evaluate the effect of this simplification, a simulation was performed to examine the relative fraction of ΔM_b to ΔM within the typical range of PS values (Supporting Information Figure S2A). We also studied the effect of neglecting ΔM_b on the PS estimation (see Supporting Information Figure S2B). It was found that neglecting ΔM_b resulted in an underestimation of PS in high PS cases, but the size of the effect was relatively small.

2.1.2 |. MRI experiment

Ten young healthy volunteers were studied (age: 24.8 ± 4.0 years, 5 males/5 females). The Single-delay WEPCAST scan used the following parameters: labeling duration = 2000 ms; PLD = 3500 ms (chosen based on an average δ_v of 4310 ms to SSS in the original report¹⁹); single-shot EPI readout; TE = 13 ms; TR = 7143 ms; encoding velocity (V_enc) = 15 cm/s; FOV = 200 × 200 mm; single slice at the mid-sagittal plane (Figure 1B); slice thickness = 10 mm; matrix = 64 × 64; voxel size = 3.13 × 3.13 mm²; parallel imaging acceleration factor = 2; flip angle = 90°; number of control/label pairs = 10; and scan duration = 4 min 49 s. A M_0,blood scan with the same TE and V_enc but a long TR of 10 s was also acquired for normalization. For comparison, a multi-PLD WEPCAST was also performed, including 3 additional PLD times (3000, 4000, and 4500 ms), with TR values of 6643, 7643, and 8143 ms; and with scan duration of 4 min 28 s, 5 min 9 s, 5 min 30 s, respectively. Thus, a multi-PLD WEPCAST data set takes approximately 20 min to acquire.

2.1.3 |. Data analysis

All MRI data were processed using in-house MatLab (version R2016a, MathWorks, Natick, MA) scripts. In WEPCAST MRI, images were acquired under 4 different conditions consisting of the complete combinations of 2 label/control conditions and 2 velocity-encoding conditions. Paired subtraction of velocity-encoded images yielded vessel-only images without tissue signal, similar to a phase-contrast MRI scan. Next, another subtraction was performed for the vessel-only images collected under control and label conditions, yielding the final WEPCAST images, that is, arterially labeled signals that remained in the vein.

Next, we sought to determine where on the SSS the δ_v equals $PLD+\frac{\tau }{2}$. Based on the theory described above, a mask encompassing the entire SSS was drawn manually on the velocity-encoded control image, which has an excellent contrast between tissue and vessel. Then, a fast-marching algorithm was applied to identify the skeleton of the vessel on the sagittal images.⁴² The skeleton was evenly divided by nodes that are 6 voxels (ie, 18.7 mm) apart and, for each node, the 20 nearest voxels within the SSS formed a region of interest (ROI) corresponding to that node. WEPCAST signals of voxels in each ROI were averaged. To account for vessel size, flow velocity, and other potential sources of signal intensity variations such as coil sensitivity, the signals were normalized by the ROI-specific M_0,blood value calculated from the corresponding M_0,blood image. The signal was further corrected by the $\frac{\frac{l}{V}}{e^{T_{1,\mathit{\text{blood}}}}}$ terms as described in Equation (7), where the distance l between nodes was obtained from fast-marching algorithm; V was assumed to be 13 cm/s⁴³; and T_1,blood was assumed to be 1584 ms.⁴⁴ The peak signal among the ROIs was then identified and considered to have a δ_v of $PLD+\frac{\tau }{2}$. The signal at this location was used to calculate E using Equation (3), with the assumptions of α = 86%.⁴⁵

For comparison, E was also estimated from the multi-delay WEPCAST data following procedures described previously,¹⁹ with the following additional assumptions: λ = 0.9 mL/g,³⁸ and T_1,tissue = 1165 ms⁴⁶; CBF in study I was based on an assumed value of 55 mL/100 g/min for males and 62 mL/100 g/min for females.¹⁹ It should be pointed out that, because the CBF term only appears in ΔM_b and not the larger dominant ΔM_a, the estimation of E is not strongly dependent on the CBF assumption. Three ROIs corresponding to anterior, middle, and posterior segments of the SSS were analyzed, and their E values were averaged. Consistency of E obtained from 2 methods was evaluated using Pearson correlation.

Finally, considering the assumptions used in analysis, including SSS velocity, T_1,blood, dispersion (σ), and CBF, we conducted sensitivity analysis to test to what extent the E estimation varies when altering these variable values by ±10%. We also tested the dependence of the method on ROI choice by using different ROI sizes (15, 20, and 25 voxels) and different ROI numbers (3, 6, and 9 voxel intervals between ROIs).

2.2 |. Study II: Assessment of intrasession, intersession, and inter-vendor reproducibility

2.2.1 |. MRI experiment

Shortening of the scan duration from 20 min to 5 min afforded the feasibility to conduct multiple measurements in 1 subject. Therefore, to evaluate the reliability of the Single-delay WEPCAST technique, we characterized its intrasession, intersession, and inter-vendor test–retest reproducibility. The sequence was implemented on two 3 Tesla platforms: a Philips Ingenia system (Ingenia, Best, the Netherlands) and a Siemens Prisma system (Prisma, Erlangen, Germany). For the spin labeling module, both implementations use the pseudo-continuous ASL approach with same labeling pulse, selective gradient, and average gradient strength. The background-suppression pulses were also carefully matched in terms of pulse shape and pulse duration. Identical imaging parameters were used on both scanners: single-shot EPI readout with TE = 13 ms; TR = 7200 ms; labeling duration = 2000 ms; PLD = 3500 ms; V_enc = 15 cm/s; FOV = 200 × 200 mm; single slice; matrix = 64 × 64; voxel size = 3.13 × 3.13 mm²; slice thickness = 10 mm; parallel imaging acceleration factor = 2; flip angle = 90°; number of control/label pairs = 10; scan duration = 4 min 55 s.

Ten healthy volunteers (age: 24.6 ± 2.3 years, 5 males/5 females) were studied. Each participant was scanned on both systems on the same day within a period of 2 hours. The order of Philips and Siemens scanner was counterbalanced across subjects. Both systems used a 32-channel head coil. The procedure of the experiment is illustrated in Figure 2. On each scanner, the subject went through 2 identical sessions with a short break and reposition between sessions, which allowed a quantification of intersession reproducibility of the technique. Within each session, WEPCAST scan was performed twice, which gave an evaluation of the intrasession reproducibility. After each WEPCAST scan, an additional M_0,blood image was acquired for normalization. Phase-contrast (PC) MRI at 4 main feeding arteries (left/right internal carotid arteries and left/right vertebral arteries)⁴⁷ were also performed for global CBF quantification. Before session 1, a T₁-MPRAGE structural scan was acquired for brain volume measures. See Supporting Information Supporting Methods for detailed imaging parameters of these sequences.

FIGURE 2.

Experimental design of the cross-vendor reproducibility study. (A) Each subject went through 2 sessions on each scanner, with a short break and repositioning between 2 sessions. The subject was scanned on scanner 1 first and then escorted to scanner 2 to repeat the same procedure as in scanner 1. Before session 1, a T₁-MPRAGE was also acquired. (B) Each session began with a TOF-angiogram. Then, WEPCAST MRI and PC-MRI for 4 vessels was performed twice. PC, phase contrast

2.2.2 |. Data analysis

For each subject, 8 measures of E, CBF, and PS were obtained (4 from each scanner) using the method described in study I. The analysis of PC MRI was described in Supporting Information. For each parameter (E, CBF, or PS), the following statistical analysis were performed: Repeated measures analysis of variance with Greenhouse-Geisser correction for sphericity violation was performed to examine whether the parameter was different between runs within 1 session, between sessions, or between scanners. Correspondence of the parameter measured on intrasession, intersession, and inter-vendor was evaluated with interclass correlation coefficient based on a single-rating 2-way mixed-effects model.⁴⁸ Bland-Altman plot was used to assess the dependence of difference on the parameter itself. Test–retest reproducibility was evaluated by coefficient of variation (CoV), including intrasession CoV, intersession CoV, and inter-vendor CoV. A paired t test was used to compare the intrasession and intersession CoV between scanners.

Finally, because the WEPCAST processing involved a manual step of drawing a mask around the SSS, we also tested the interrater variability by having 2 raters analyze the same data sets independently. A dice index was calculated based on the overlapping voxel number of the masks divided by the average voxel number. Interclass correlation coefficient was used to determine the dependence of the PS results on raters.

2.3 |. Study III: Comparison of WEPCAST MRI with contrast agent–based method

2.3.1 |. MRI experiments

Although there has not been a gold standard MRI method to measure BBB permeability to water, several contrast agent–based approaches have been suggested to be sensitive to this index based on the notion that faster water exchange will result in more pronounced T₁ relaxivity in the tissue.^33–35,49 Therefore, in this substudy, we compared WEPCAST to a contrast agent–based method reported in Uh et al.⁴⁹ Nine healthy volunteers were enrolled (age: 34.0 ± 11.3 years, 2 males/7 females). Each subject underwent a noncontrast WEPCAST MRI followed by a dynamic contrast agent–based MRI scan.

WEPCAST MRI was conducted with the same parameters as in study II, except that TR = 7050 ms; TE = 14 ms; and scan duration = 4 min 49 s. An M_0,blood image was acquired for normalization. PC MRI and a T₁-MPRAGE were also acquired for CBF measurement. The parameters of PC and MPRAGE scans were identical to those used in study II.

After WEPCAST MRI, the contrast agent method was performed, which consisted of an M₀ scan and 30 T₁-weighted images acquired with a multiple spin-echo inversion recovery sequence known as the vascular space occupancy (VASO) sequence.⁴⁹ The M₀ scan used identical acquisition parameters as the VASO, but no inversion pulse was played out and the TR was 10 s. Figure 3A illustrates the scan procedure. The imaging parameters were as follows: 8-echo spin echo with an echo interval (τ_CPMG) of 10 ms; TR = 6000 ms; TI = 1088 ms; FOV = 230 × 230 mm²; scan matrix = 128 × 128; single slice with slice thickness = 5 mm; EPI factor = 7; SENSE factor = 2; and duration = 54 s. The total scan time was 27.3 min. The use of multi-echo acquisition allowed the separation of T₁ and T₂ relaxivity effects of the contrast agent because we are primarily interested in the T₁ relaxivity in permeability assessments. Contrast agent (gadoteridol) was administrated intravenously at a dosage of 0.1 mmol/kg at the beginning of the 4th dynamic with a rate of 5 mL/s via a power injector (Figure 3A).

FIGURE 3.

Contrast agent–based method for the measurement of BBB permeability to water. (A) Experimental protocol. A total of 30 vascular space occupancy (VASO) MRI scans were performed at the timing shown: 3 scans before contrast agent injection and 27 immediately after. Each VASO scan used an inversion recovery multi-spin echo sequence. The multi-echo data can be extrapolated to a TE = 0 ms, thereby minimizing T₂ effect of Gd. (B) Two compartment water exchange model used for VASO signal fitting. K_in and K_out are exchange rates of water between blood and tissue. The exchange rate is affected by BBB permeability (PS) and cerebral blood volume (CBV). λ is brain–blood partition coefficient for water. CBV, cerebral blood volume; PS, permeability surface area product; VASO, vascular space occupancy

2.3.2 |. Data analysis

WEPCAST and PC MRI data were processed following the procedure described in study I and II. For contrast-agent data, quantitative analysis focused on gray matter because it has a higher vessel density and thus signal reliability. The gray matter ROI was obtained by segmenting the MPRAGE image using SPM12 (Wellcome Trust, University College London, UK), and the signals of the voxels within the ROI were averaged. The multi-echo ROI data were first fitted as a function of TE to obtain T₂ and S₀ time courses. Next, the initial concentration of Gd in the gray matter ROI ([Gd]) was estimated for each participant from the T₂ difference between precontrast and first postcontrast images based on

$$\frac{1}{T_{2,pre}}-\frac{1}{T_{2,\mathit{\text{post}},0}}=J\cdot {[Gd]}_0,$$ (8)

where J is the T₂ relaxivity of the Gd contrast agent (assumed to be 5 s⁻¹ mM⁻¹).⁴⁹ Then, [Gd]₀ was averaged across all participants to improve reliability and was used as the initial Gd concentration in all subjects.

The Gd concentration decay was modeled as a biexponential decay so that the Gd concentration in blood can be written as

$${[Gd]}_b(t)=a_1{e}^{-n_{1}t}+a_2{e}^{-n_{2}t},$$ (9)

with ${[Gd]}_b(t=0.9\text{min})=\frac{{[Gd]}_0}{CBV}$, where n₁ and n₂ are gadolinium decay rates; a₁ and a₂ are coefficients associated with the 2 components; and CBV is cerebral blood volume. We also assumed that n₂ = 0.0434*n₁ and a₂ = 0.5*a₁ based on the work of Uh et al.⁴⁹

The [Gd]_b (t) time course is in turn related to the S₀ time course through a 2-compartment water exchange model (see Supporting Information), which is experimentally measured (ie, the VASO S₀ signal). In this fitting, the unknowns are the gray matter PS (PS_VASO,gm), CBV, and Gd decay rate (n₁) (Figure 3B).^35,50 Details of the model and the fitting procedures are described in Supporting Information.

To facilitate the comparison of the contrast-agent results with that of WEPCAST MRI, which is a global measurement and reflects an average of gray and white matter PS, PS_VASO,gm was converted to PS_VASO,global using:

$$P{S}_{\mathit{\text{VASO}},\mathit{\text{global}}}=P{S}_{\mathit{\text{VASO}},gm}\cdot \frac{V_{gm}}{Vgm+Vwm}+R\cdot P{S}_{\mathit{\text{VASO}},gm}\cdot \frac{V_{wm}}{Vgm+Vwm},$$ (10)

where V_gm and V_wm are volumes of gray and white matter measured from the individual’s MPRAGE image, and R is the ratio of white matter and gray matter PS and was assumed to be 0.27 based on the CBV ratio.⁵¹ Pearson correlation was then used to evaluate the consistency of PS obtained from the 2 methods (PS_WEPCAST and PS_VASO,global).

3 |. RESULTS

3.1 |. Study I: Development of Single-delay WEPCAST MRI

Figure 4A shows representative data from a single-delay WEPCAST scan, including control, label, and difference images. As can be seen, due to the phase-contrast flow-encoded acquisition scheme, vessel signals are selectively measured and tissue signals are successfully suppressed. A spatial profile of the normalized WEPCAST signals along the SSS is shown in Figure 4B. A clear peak can be seen. As described in the Methods section, this location is then considered to have a δ_v of 4500 ms, and its signal is used to estimate E. Across the whole group (N = 10), the water extraction fraction E was found to be 92.2 ± 2.8%.

FIGURE 4.

Representative images of WEPCAST MRI and relationship between single- and multi-delay results. (A) Single-delay WEPCAST control, label, and difference images from a representative subject. (B) WEPCAST signal profile along SSS. (C) Scatter plot of E obtained from Single-delay WEPCAST and multi-delay WEPCAST across all participants (N = 10). SSS, superior sagittal sinus

Next, we evaluated the dependence of E estimates on model assumptions as well as on ROI sizes and numbers. The results are shown in Table 1. To study the error propagations, we also included the PS values. As can be seen, the parameter assumptions have relatively small effects on E and PS estimates. Blood T₁ seemed to have the greatest effect on the estimation.

TABLE 1.

Dependence of estimated water extraction fraction and permeability surface area product on model assumptions and ROI choices region

	E (%)	PS (mL/100 g/min)
Main text	92.2 ± 2.8	152.5 ± 18.7
T1,blood − 10%	89.5 ± 4.0	136.6 ± 25.4
T1,blood + 10%	93.8 ± 2.2	166.3 ± 18.6
SSS velocity − 10%	92.3 ± 2.9	155.0 ± 25.0
SSS velocity + 10%	92.0 ± 2.9	152.6 ± 18.8
σ − 10%	92.4 ± 2.7	153.7 ± 18.7
σ + 10%	92.0 ± 2.8	151.1 ± 18.8
CBF −10%	92.2 ± 2.8	137.3 ± 16.9
CBF+ 10%	92.2 ± 2.8	167.8 ± 20.6
ROI size = 15 voxels	91.6 ± 2.6	147.5 ± 17.6
ROI size = 25 voxels	92.6 ± 3.0	157.7 ± 25.2
ROI interval = 3 voxel long	92.3 ± 2.9	153.0 ± 18.9
ROI interval = 9 voxel long	92.3 ± 2.6	152.6 ± 17.8

CBF, cerebral blood flow; E, extraction fraction; PS, permeability surface area product; ROI, region of interest; SSS, superior sagittal sinus.

Analysis of the multi-delay WEPCAST data (with a scan time of 20 min) yielded an E of 94.0 ± 2.0%. Figure 4C shows a scatter plots between the Single-delay and multi-delay results. A strong correlation between results from the 2 methods is seen (R = 0.82 and P = .004), suggesting that the Single-delay method can provide comparable results to that of multi-delay while reducing the scan time by 4 times to a clinically applicable duration of approximately 5 min.

3.2 |. Study II: Assessment of intrasession, intersession, and inter-vendor reproducibility

Figure 5A displays WEPCAST control, label, and difference images of a representative participant from both Philips and Siemens scanners. Figure 5B shows the corresponding WEPCAST signals along the SSS. Table 2 summarizes the PS values across all participants in all measurements. Analysis of variance analysis revealed no differences in E, CBF, or PS between intrasession, intersession, or inter-vendor results (P = .94, 0.57 and 0.40 for E; P = .45, 0.75 and 0.06 for CBF; P = .59, 0.78 and 0.47 for PS).

FIGURE 5.

Test–retest reproducibility of WEPCAST MRI. (A) WEPCAST control, label, and difference images of 1 subject from 2 scanners. (B) WEPCAST signal profile along superior sagittal sinus of the subjects for 2 scanners. (C) Intrasession CoV of PS. (D) Intersession CoV of PS. (E) Inter-scanner CoV of PS. CoV, coefficient of variation

TABLE 2.

Summary of PS (mL/100 g/min) results from all participants (Mean ± SE, N = 10)

	Session 1		Session 2
	Run 1	Run 2	Run 1	Run 2
Philips (Best, the Netherlands)	133.7 ± 5.3	143.4 ± 6.0	138.1 ± 4.2	141.6 ± 4.5
Siemens (Erlangen, Germany)	136.0 ± 6.0	135.4 ± 6.5	138.3 ± 3.6	133.6 ± 4.2

SE, standard error.

Figure 5C through E display the intrasession, intersession, and inter-vendor CoV for PS. It can be seen that the intrasession and intersession CoV on both scanners were between 5% and 8%, suggesting a good reliability of the technique. The CoV values obtained from 2 scanners were not significantly different (P = .58 for intra-session; P = .65 for intersession). The inter-vendor CoV was 8.9 ± 3.0%, indicating that different scanners may yield additional variations. The intersubject CoV in PS was also computed and found to be 6.7%. The corresponding CoV values for E and CBF, which were used to calculate PS, are summarized in Supporting Information Table S1. As can be seen, they had lower CoV compared to PS. As an additional observation, we found that there was a significant correlation between E and CBF across participants. Individuals with a higher CBF tended to have a lower E (Supporting Information Figure S3) (R = −0.81, P = .005).

Interclass correlation coefficient results of E, CBF, and PS are summarized in Table 3. A significant correlation was observed in all comparisons, except for inter-vendor PS values. Figure 6 shows scatter plots of these measures for intra-session results. Bland-Altman plots comparing these measures for intrasession, intersession, and inter-vendor results are shown in Supporting Information Figure S4.

TABLE 3.

Summary of intrasession, intersession, and inter-vendor ICC results

		ICC	CI	P Value
E	Intrasession	0.53	[0.27, 0.72]	<.001
	Intersession	0.72	[0.41, 0.88]	<.001
	Inter-vendor	0.76	[0.30, 0.94]	.003
CBF	Intrasession	0.88	[0.78, 0.93]	<.001
	Intersession	0.88	[0.72, 0.95]	<.001
	Inter-vendor	0.84	[0.49, 0.96]	<.001
PS	Intrasession	0.37	[0.08, 0.61]	.008
	Intersession	0.48	[0.06, 0.76]	.014
	Inter-vendor	0.26	[−0.41, 0.74]	.22

CI, confidence interval; ICC, interclass correlation coefficient.

FIGURE 6.

Scatter plots of intrasession reproducibility data. (A) Scatter plot of E from 2 runs. (B) Scatter plot of CBF from 2 runs. (C) Scatter plot of PS from 2 runs. CBF, cerebral blood flow

Finally, interrater reliability of the results was also examined using data from all repetitions for all participants (ie, 80 data sets in total). The dice index of the ROIs drawn by 2 independent raters was 0.86 ± 0.04. Scatter plots of the PS values obtained by the raters are shown in Supporting Information Figure S5, demonstrating good consistency (interclass correlation coefficient = 0.90 with 95% C.I. [0.85, 0.93], P < .0001).

3.3 |. Study III: Comparison of WEPCAST MRI with contrast agent–based method

From WEPCAST MRI, E was found to be 91.2 ± 3.8% (mean ± SD, N = 9). Global CBF measured with PC MRI was 55.8 ± 9.2 mL/100 g/min. PS_WEPCAST was then estimated to be 138.1 ± 26.2 mL/100 g/min.

PS was also measured with contrast agent–based method. Figure 7A shows dynamic VASO images for pre- and postcontrast agent conditions as a function of time. The time course of gray matter S₀ signal is shown in Figure 7B. It can be seen that, following the Gd injection, the VASO signal revealed an immediate increase due to the effects of the contrast agent on intravascular (CBV-related) and extravascular (water exchange–related) tissue. Fitting the signal time course to the kinetic model yielded a PS_VASO,global of 161.0 ± 62.5 mL/100 g/min (N = 9) with a CBV of 3.7 ± 0.6% and Gd decay time constants of 1.0 ± 0.5 min and 23.9 ± 11.1 min. Figure 7C shows a scatter plot between PS_WEPCAST and PS_VASO,global. It can be seen that the PS obtained from WEPCAST MRI showed a significant correlation with that from the contrast agent–based method (R = 0.73, P = .02).

FIGURE 7.

Results of contrast agent–based BBB method in comparison with Single-delay WEPCAST. (A) Representative VASO images (first echo) before and after contrast agent injection. (B) Gray matter VASO signal as a function of time and corresponding fitting result from 1 subject. (C) Scatter plot of PS results from WEPCAST and VASO MRI across all participants (N = 9)

4 |. DISCUSSION

Noncontrast assessment of BBB permeability to water represents a new area of research and may become useful biomarkers in several diseases; however, the current methods often require a long scan time. The present work proposed a new, Single-delay WEPCAST method that can provide an estimation of BBB permeability within a scan time of approximately 5 min. Reliability of the method was evaluated in terms of intrasession, intersession, and inter-vendor test–retest reproducibility, and this revealed a CoV range of 6% to 9%. Although there are no gold standard MRI techniques for water permeability measurement, some contrast-agent methods have been suggested to be sensitive to this parameter. Therefore, we also compared the non-contrast WEPCAST technique with a contrast agent–based method and observed a significant correlation in results from these 2 methods.

Abnormalities in blood–brain barrier permeability have been implicated in a large number of diseases, such as Alzheimer disease,² multiple sclerosis,³ traumatic brain injury,⁵² sickle cell anemia,⁵³ and cerebrovascular diseases.^4,54 Existing BBB permeability techniques suitable for humans are largely based on the measurement of permeability to gadolinium using dynamic contrast-enhanced MRI⁶ or permeability to albumin using lumbar puncture.⁵⁵ Recently there has been a surging interest in the measurement of water permeability using MRI. Such techniques have the advantages that they are completely noninvasive, and because water is a small molecule, they may be sensitive to early BBB disruptions. The basic principle of such methods is to use water as an endogenous tracer (through spin labeling) and aims to differentiate blood water molecules that are exchanged at the capillary–tissue interface from those remaining in the vessel. The relative fractions of these 2 spin pools can then be used to obtain a quantification of BBB permeability to water, following some kinetic modeling that accounts for T₁ relaxation, bolus arrival time, post-labeling delay, and so forth. One method to distinguish the intravascular and extravascular spins is based on the diffusion properties of the water, in which spins in the capillary network experiences a much greater apparent diffusion coefficient following the intravascular incoherent motion principle.^{13,18,26–30,32} Another method exploited the T₂ differences between tissue and blood and used multi-TE acquisition in combination of biexponential fitting to determine the fractions of intravascular and extravascular spins.^22–25 Other methods utilizing T₁ difference or magnetization transfer effect between intravascular and extravascular compartments were also proposed.^17,20,21,56 The WEPCAST technique takes advantage of the differences in spatial destiny between exchanged and non-exchanged spins and measures the signals in large draining veins. The acquisition module of the sequence was also devised to be velocity-selective such that extravascular signal was further suppressed. Much technical work is still needed on each of the above-mentioned approaches to provide an efficient and reliable method ready for routine clinical use. The present work represents an effort to advance the WEPCAST method toward this general goal.

As in other methods, the parameter we want to determine in WEPCAST, that is, unexchanged water fraction, is tangled with other parameters such as bolus arrival time in terms of their effects on the MR signal. This necessitates multiple experimental measures in order to tease apart these parameters. In the proof-of-principle report of WEPCAST MRI, this was achieved by performing the experiments at multiple PLD times, which resulted in a long scan time of approximately 20 min. The conceptual advance in the present method is that we utilized the spatial profile of the signal along the SSS and exploited the spatiotemporal relationship of flowing spins to transform the model into a function of space rather than time, allowing the estimation of E using a single-delay scan of 5 min. In terms of the choice of the Single-delay time, the guiding principle is that the peak of the WEPCAST signal (eg, Figure 1D) should be located on the SSS. That is, the delay time should not be too short that the labeled bolus has not reached the SSS, and it should not be too long that the bolus has left the vein. Because the peak of the signal occurs at $PLD+\frac{\tau }{2}$, the PLD should be chosen to be approximately ${\delta }_v-\frac{\tau }{2}$, where δ_v is the range of bolus arrival time on the SSS. Previous reports have suggested that δ_v for posterior SSS is about 4810 ms.¹⁹ Thus, a PLD of 3500 ms was used in the present study. It is also expected that, for older participants, a longer PLD should be used.

Test–retest reproducibility is a critical index in the assessment of a technique in terms of its potential in clinical applications. This study showed that PS measured with the proposed technique has an intrasession and intersession CoV of around 6%, which is comparable or slightly lower than that of ASL (around 10%).^57–59 In the context of pathological alterations in BBB permeability, it was reported that BBB permeability to water can change by 28.2% in diabetes,¹³ 19.5% in hypercholesterolemia,¹³ 31.0% in multiple sclerosis,³⁴ and 93.8% in brain tumor.³⁴ It has also been reported that BBB permeability to water could increase by 32% in aging mice⁶⁰ and 50% to 100% in rat models of Alzheimer disease.³³ Thus, the reliability of the WEPCAST measure is expected to be sufficient for detection of the BBB abnormalities in diseases. On the contrary, because the WEPCAST signal is proportional to 1-E and the 1-E term in healthy subjects is only ~8% on average, the dynamic range for 1-E to further decrease in disease (when PS increases) may be limited. This also means that CoV of PS in patients may be higher than that observed in healthy subjects. To prepare the technique for multi-site studies, we further harmonized the sequence across MRI vendors and found an inter-vendor CoV of 8.4%, suggesting that the technique has a good reproducibility across MRI platforms.

There is not a gold standard method for water permeability measurement in humans; therefore, validation of the proposed technique has not been feasible. In the literature of contrast-agent MRI, on the contrary, a few reports have suggested the sensitivity of T₁-weighted signal to water permeability.^33–35,49 With these frameworks, dynamic acquisition of T₁-weighted MRI is performed before and after Gd contrast-agent injection. Whereas the signal change is primarily indicative of Gd leakage, thereby BBB permeability to Gd, the MR signal can also be shown to be sensitive to water exchange across the capillary–tissue interface under the fast-exchange regime. Although this effect is relatively small and intensive research is still ongoing to separate it from Gd leakage and CBV effects, it is nonetheless informative to compare the WEPCAST method to such contrast agent–based techniques. The results revealed a significant correlation between PS values measured from the 2 approaches.

The current study has several limitations: First, although the single-delay WEPCAST technique reduced the scan time by 75%, it also required more assumptions, data analysis steps, and a manual drawing of ROI. These additional considerations may present confounding factors. Two of the assumptions used were blood T₁ and pseudo-continuous ASL labeling efficiency, which may be experimentally determined with additional sequences.^39,45,61 The venous flow velocity along the SSS could also be measured with a separate phase-contrast scan using a thin slice. Second, only 2 MRI vendors, that is, Philips and Siemens, were included in the study. Further effort will be needed to implement the techniques on other platforms. Third, whereas the comparison with contrast-agent method provides some corroboration of the proposed technique, it has not been compared with a gold standard technique. Future study should compare the WEPCAST technique to other methods, for example, [¹¹C]Butanol positron emission tomography,¹⁵ and apply experimental techniques that are able to transiently modulate BBB permeability. Finally, WEPCAST MRI is a global technique but does not provide regional information of BBB permeability in the present implementation. Therefore, compared to regional BBB techniques proposed in the literature,^{13,24–26,28,29,33,60} this method may not be sensitive to diseases that only affect a small region of the brain such as brain tumor or that detect early focal changes occurring in small structures such as hippocampus in early Alzheimer disease. Future studies shall focus on the measurement of WEPCAST signal in smaller draining veins such as internal cerebral veins and straight sinus, which may provide region-specific information of BBB integrity.

5 |. CONCLUSION

In this study, we proposed a time-efficient, Single-delay WEPCAST technique that can provide an assessment of global BBB permeability to water within a scan time of 5 min. We determined the benchmarks of this technique in terms of intrasession, intersession, and inter-vendor test–retest reproducibility. The results of the noncontrast method were also corroborated by a contrast agent–based method. Single-delay WEPCAST MRI may provide a useful tool for the assessment of BBB breakdown in neurologic and psychiatric diseases.

Supplementary Material

Supplementary Material

FIGURE S1 Simulated plots of function d_a(t, δ_v) in Equation (5) for different location (l = −10, 0, 10 cm) at superior sagittal sinus (SSS) with labeling duration (τ) = 2s and post-labeling delay (PLD) = 3.5s. Location l = 0 cm has a bolus arrival time of $\text{PLD}+\frac{\tau }{2}$. Red line denotes time of acquisition t = PLD + τ, ie, 5.5s in this case

FIGURE S2 Simulation on the effect of neglecting re-exchanged spins on PS estimation. (A) Relative fraction of re-exchanged spin signal (ΔM_b) to the total WEPCAST signal as a function of PS. The PS range was selected to be mean ± 2 standard deviations of values reported in previous PET literature. (B) Estimated PS when neglecting the contribution of ΔM_b as a function of true PS

FIGURE S3 Scatter plot between E and CBF across participants (N = 10). Individuals who have a higher CBF had a lower E

FIGURE S4 Bland-Altman plots of E, CBF and PS for intra-session, inter-session and inter-vendor comparisons

FIGURE S5 Scatter plot of the PS values obtained from two independent raters

TABLE S1 CoV (%) results of E and CBF (N = 10) from the reproducibility study