Blood–Brain Barrier Permeability to Water Measured Using Multiple Echo Time Arterial Spin Labeling MRI in the Aging Human Brain

The blood–brain barrier (BBB) plays a vital role in maintaining brain homeostasis, but the integrity of this barrier deteriorates slowly with aging. Noninvasive water exchange magnetic resonance imaging (MRI) methods may identify changes in the BBB occurring with healthy aging.

function of the BBB.Dynamic contrast-enhanced (DCE) MRI, which makes use of Gadolinium-based contrast agents (GBCA), are widely used to access the vessel permeability in the brain. 4nfortunately, due to the large molecular size of GBCA, this technique is only suited for detecting major disruptions in the BBB. 5 Novel and emerging methods are being developed to detect early changes occurring in BBB dysfunction using smaller sized molecules.Among these techniques, arterial spin labeling (ASL)-based methods for water exchange estimation are gaining popularity because of the use of water within blood as an endogenous tracer. 6Different properties of labeled water, such as transverse relaxation [7][8][9][10][11] and diffusion coefficient [12][13][14] are exploited to compartmentalize it in intravascular blood (IV) and extravascular tissue (EV) compartments, and the transition between these two compartments is proposed as a measure of BBB permeability.
Movement of water through the brain plasma membrane takes place through different mechanisms including a very slow process of simple diffusion, by aquaporins and by cotransport with organic or inorganic ions. 15Changes in various transport systems in the brain are reported to occur with healthy aging, 16 although water exchange systems have not been investigated thoroughly.A multiple echo time (multi-TE) ASL-based study reported a decrease in the rate of water flux in aquaporin deficient mice compared to normal mice. 17n another study, using a similar method, authors reported an increased BBB permeability in aged mice along with an increase in mRNA expression of aquaporins and a decrease in RNA expression of α-syntrophin-a protein responsible for anchoring aquaporin-4 to the BBB. 9 This shows a complex interplay of underlying molecular processes taking place during healthy aging.Conversely, a recent study using diffusionweighted ASL (DW-ASL) reported reduced BBB permeability with age in healthy humans. 12The authors argued that this opposing trend of reduced BBB permeability with age in humans compared to the multi-TE study in mice may have arisen due to interspecies differences.This highlights a need to investigate age-based changes occurring in water transport mechanisms in the brain, as an index of BBB permeability, and to further unravel the overlaps and differences in BBB modularity assessed with different methods, and between healthy aging and neurodegenerative diseases.
This study aimed to apply multi-TE ASL method to assess age-related differences in BBB permeability to water in healthy humans.Furthermore, two model approaches of variable complexity were applied to establish whether these changes in BBB permeability can be detected using a noninvasive MRI method.

Study Population
All experiments in this study were approved by the Ethical Committee of the University of Bremen, Bremen, Germany.Written informed consent was obtained before the participants were enrolled in the study.
Data were acquired from two different age groups of healthy volunteers-older group (≥50 years) and younger group (≤20 years) who fulfilled the inclusion criteria of having no previous medical history of neurological or neurodegenerative diseases.All volunteers were examined at 3T (MAGNETOM Skyra, Siemens Healthineers AG) using a 20-channel head coil.

Data Analysis: Physiologically Informed Biophysical Model
In the first approach, an extended multi-TE two compartment model 8 was applied to estimate exchange time T ex ð Þ as a proxy measure for BBB permeability.This approach, exhibiting higher complexity, is referred here as the "Physiologically Informed Biophysical (PIB)" model.
The measured pCASL signal is assumed to be composed of three components.When the signal arrives in the imaging voxel at t > ATT, the signal coming from the labeled spins in the arteries forms the first component S bl1 ð Þ.At t > ATT þ ITT, the labeled spins after traversing through the arteries and arterioles reach the capillary exchange site, where the signal within capillaries constitute the intravascular component S bl2 ð Þ and the signal coming from the tissue after exchange resamples the extravascular component S ex ð Þ.The equations for each component are given below (for details see Mahroo et al 8 ): where M 0 is the arterial longitudinal equilibrium magnetization, f is perfusion, c t ð Þ is an arterial input function, m bl , m ex are the magnetization relaxation function of blood and tissue, and r blex is the blood to tissue exchange function.
c t ð Þ considers four different time cases: where α is the labeling efficiency and T 1 bl is the longitudinal relaxation of the blood.r blex is defined as: where T ex describes the time taken by the labeled spins to move from blood to tissue after exchange through the BBB and acts as a proxy measure of BBB permeability.m bl t ð Þ and m ex t ð Þ are defined as: where T 1 ex is the longitudinal relaxation of the tissue.
For PIB model processing, data were analyzed with the Oxford Center for Functional MRI of the Brain (FMRIB)'s Software Library (FSL). 20Structural T1 MPRAGE images were preprocessed with fsl_anat pipeline.The encoded ASL time series were motion corrected using MCFLIRT with a six-parameter rigid transformation, and distortion corrected using M0 images acquired in phase reversed directions (RL and LR) using TOPUP.The ASL signal at each TI and TE was decoded by applying the respective Hadamard decoding matrix.The resulting decoded images from the two protocols were concatenated.The in-house developed PIB model was incorporated in the Bayesian non-linear fitting framework of FSL FABBER 21 and the final dataset was fitted voxel-wise within the whole brain.The T 1 bl , T 1 ex , T 2 bl , and T 2 ex values used in the model were 1664, 1331, 165, and 85 msec, respectively, taken from the literature. 22The resulting four fitted parameters were cerebral perfusion (CBF), arterial transit time (ATT), exchange time T ex ð Þ, and intravoxel transit time (ITT).Mean gray matter (GM) values were calculated using a 50% gray matter probability mask.The parameter maps were smoothed (posthoc) with a 2Â voxel size isotropic Gaussian kernel (FWHM) to improve signal-to-noise ratio (SNR).Finally, the parameter maps were registered to structural and Montreal Neurological Institute (MNI) 152 (2 mm) standard spaces to compare them within-and between-subjects, respectively.

Data Analysis: Triexponential Decay Model
A second approach applied a simpler method inspired by Schidlowski et al 23 where the authors applied a biexponential equation to compartmentalize labeled water into IV blood and EV tissue compartments based on difference in T2 transverse relaxation and further used a simple line equation to estimate water transition rate k lin ð Þ as a surrogate measure of BBB permeability.In the current model, the biexponential equation was extended to triexponential to additionally accommodate a third T2 component originating from cerebrospinal fluid (CSF).This simpler approach with a smaller number of assumptions is referred as the "Triexponential Decay (TD)" model.
Preprocessed HAD-4 TIs [1600, 2600, 3600] msec were used for TD modeling as these TIs were acquired with a longer SBD (1000 msec) and provided better SNR compared to HAD-8 TIs.Moreover, HAD-4 TIs spanned over a wide range covering the reported cerebral vascular water transit time of 3000-4000 msec. 24uring this time, the labeled water is assumed to be transitioning between different compartments.

Bulk ASL and M0 T2 Mapping
Preprocessed HAD-4 images were fitted using monoexponential decay function to estimate bulk whole brain T2 voxelwise using Python non-linear least squares (NLLS) curve_fit function from the Scipy.Optimize library with the following model: where S is the measured signal, S 0 is the signal amplitude, TE is the echo time, and T 2 is the transverse relaxation of the bulk signal.
The whole brain T2 maps represent contributions from all compartments at a specific TI.
For approximation of T2 EV tissue T 2 EV ð Þ, multi-TE M0 images were fitted using Eq. 8. Since M0 images were not background suppressed, it was assumed that the signal will dominantly consist of tissue contribution and the resulting fitted T2 was regarded as T 2 EV .It should be noted that, in reality, the signal contributions from other compartments will also be present in the voxel.Nevertheless, this approach may provide a subject-specific approximation of T 2 EV voxelwise as compared to using a fixed global value from the literature.

Tri-Exponential Compartmentalization
To improve fitting on low SNR ASL perfusion images, a smoothing filter with a 5Â voxel size isotropic Gaussian kernel (FWHM) was used to decrease noise.As discussed earlier, the T2 fitted with multi-TE M0 was used as EV tissue T 2 EV .On the other hand, IV blood T 2 IV and CSF T 2 CSF were assumed to be 165 and 2200 msec, respectively, taken from the literature. 22All three T2 values were taken in the model as fixed input parameters.
Preprocessed HAD-4 ASL images were fitted voxelwise with a three-parameter triexponential decay model to compartmentalize labeled water using a curve_fit NLLS approach with the following equation: where S ASL TI ð Þ is the total signal at a specific TI; S EV , S IV , and S CSF are the signal contributions of extravascular, intravascular and CSF compartments, respectively.A threshold value of 0.9 of root mean squared error (RMSE) was used to include voxels which showed a reliable fit for Eqs. 8 and 9.

Tissue Transition Rate
The resulting fitted signals were used to estimate TI dependent extravascular tissue fraction (f EV Þ given as: Similarly, intravascular blood fraction f IV at a specific TI could be calculated to examine the dynamic change based on intravascular blood signal transition.For further analysis, only f EV was used for determining compartmental transition rate.
Frequency histograms of f EV at different TIs were calculated, and peak fractions were detected.Compartmental transition rate was modeled using a linear model: The slope k lin reflects the rate of change of f EV over TI which is considered as a proxy measure of BBB permeability using this model approach.

Statistical Analysis
Mean GM values of CBF, ATT, T ex , and ITT were compared between the groups using a two-tailed unpaired Student t-test.f EV and k lin values were compared between the groups using a two-tailed unpaired Student t-test.Association between the two model approaches was investigated using Pearson's correlation coefficient between f EV and T ex , and k lin and T ex : Effect size for CBF, ATT, T ex , ITT, and k lin was calculated: effect size was defined as the difference between the groups divided by the pooled standard deviation of the groups; where by convention an effect size of 0.2, 0.5, and 0.8 are considered as small, medium, and large, respectively. 25P values <0.05 were considered statistically significant.

Results
Both groups consisted of 13 volunteers each, with the older group having a mean age of 56 AE 4 (females = 5) and the younger group having a mean age of 18 AE 1 (females = 7), demographic details can be found in Table S1 in the Supplemental Material.
Figure 1 shows a complete decoded ASL dataset (HAD-8 + HAD-4) of a representative younger volunteer (age = 20 years) at different TIs and TEs.As early as TI = 600 msec, labeled blood can be seen arriving in the bigger arteries and at TI = 2200 msec, it perfuses into the whole brain.At later TIs, signal intensity decreases mainly due to T1 relaxation and at later TEs due to T2 relaxation.A higher signal intensity of HAD-4 TIs is visible from the volunteer data which is due to the longer SBD of 1000 msec used for HAD-4 protocol as compared to HAD-8 (SBD = 400 msec).Figure 2 compares signal curves generated using HAD-8 protocol (Fig. 2a) and HAD-4 protocol (Fig. 2b) along with the representative images of a younger (20 years) and an older volunteer (53 years) at the respective sampling TIs.

Physiologically Informed Biophysical Model
The group averaged mean gray matter values and effect sizes of the fitted parameters CBF, ATT, T ex , and ITT are summarized in Table 1.It was observed that the T ex was significantly lower in the older group (143 AE 30 msec, effect size = 1.9) compared to the younger group (224 AE 51 msec; Fig. 3a).
Figure 4 shows various slices of group averaged T ex maps along with the normalized difference map of T ex between the groups.

Triexponential Decay Model
Figure 5a shows example T2 maps.In Fig. 5b, by visual inspection, it can be seen that T2 decreases as TI increases in the fitted T2 maps as well as in T2 frequency histograms.Furthermore, it can be seen that T2 could be reliably fitted in gray matter.But for the last TI, the fitting resulted in very few voxels of white matter, possibly due to lack of original ASL signal which can be seen in the example decoded ASL images shown in Fig. 1.In Fig. 5c,d, a T2 map fitted with multi-TE M0 data and its frequency histogram are shown, respectively.
In Fig. 6, compartmental transition of labeled spins in terms of EV tissue fraction f EV and IV blood fraction f IV À Á is shown.A trend of decrease in f IV over TIs and increase in f EV can be seen.A threshold of RMSE >0.9 was used from triexponential fit (Eq.9) for selecting voxels with a reliable fit and therefore, white matter voxels in general and many voxels at TI = 3600 msec appear to be missing.Changes in f EV dynamics over increasing TIs are shown in Fig. 7b representing a histogram and peak frequencies.For this volunteer, the peak maximum frequencies for TIs = 1600, 2600, and 3600 msec were 0.52, 0.72, and 0.84, respectively.The compartmental transition rate k lin for the volunteer was found to be 0.161 s À1 (Fig. 7c).
A comparison of f EV between the two groups is shown in Fig. 8a.In the younger group, f EV gradually increased over the course of TIs and the group averaged f EV at TIs = 1600,  2600, and 3600 msec were 0.54 AE 0.07, 0.69 AE 0.12, and 0.83 AE 0.06, respectively.Conversely, f EV in the older group at TI = 1600 msec was significantly higher (0.70 AE 0.09) than the younger group, which then decreased slightly to 0.66 AE 0.09 at TI = 2600 and increased again to 0.81 AE 0.07 with no significant differences (P = 0.41 and 0.38, respectively) in comparison to the younger group.The group averaged k lin value in the younger group was significantly higher (0.150 AE 0.038 s À1 , effect size = 2.6) than the older group (0.054 AE 0.039 s À1 ; Fig. 8b).The group averaged k lin slopes for the two groups along with the individual peak frequencies are shown in Fig. 8c.The slope of the younger group was steeper than the older group and the major difference seems to appear from the higher f EV at TI = 1600 msec in the older group as compared to the younger group.k lin can similarly be calculated for change in dynamics of IV blood fraction (f IV ), results shown in Fig. S1 in the Supplemental Material.When comparing the two model approaches, a significant negative correlation (r = À0.80) between f EV at TI = 1600 msec and T ex was found (Fig. 9a).While, the correlation between f EV at TI = 2600 msec and T ex ; and f EV at TI = 3600 msec and T ex was very weak and not significant (r = 0.18, P = 0.39 and r = 0.06, P = 0.78, respectively; Fig. 9b,c).Correlation between k lin and T ex was significantly positive (r = 0.73; Fig. 9d).

Discussion
In this work, multi-TE ASL imaging was applied to detect changes in permeability of the BBB occurring due to healthy aging.Two approaches of variable complexity were investigated, where one allowed for physiologically informed estimation of exchange time (T ex Þ and the other applied a simple triexponential decay to measure compartmental transition rate (k lin Þ.Both approaches were able to detect age-based changes and provided significant differences between the two groups.The PIB model approach showed lower exchange time in the older group, depicting a less hindered movement of water across the BBB.The TD model approach revealed lower k lin and a flatter slope for the older group compared to the younger one.This may provide evidence that multi-TE ASL method is able to detect age-based changes in BBB permeability.
The PIB model is based on the well-known general kinetic model proposed by Buxton et al. 26 It was further extended to a two-compartment model and T2 decay was introduced to compartmentalize the labeled blood by probing T2 transverse relaxation property. 7Recently, it was further modified to include intravoxel transit time to incorporate transit through the vessels, within the voxel, before the labeled water reaches the capillary exchange site. 8This provides us with a physiologically well-informed model where various properties of water like T1 decay, T2 decay and transit times are considered to make robust estimations.Furthermore, Bayesian inference 21 was applied to include physiologically plausible prior values for parameter estimation which assisted in fitting noisy ASL data and provided interpretable parameter maps.This model considers many parameters and makes a number of assumptions as the bolus travels through the brain vasculature.These assumptions, on one hand, equip the model to explain the biophysical and physiological mechanisms but, on the other hand, also increase the complexity of the model.Furthermore, these assumptions may limit the model analysis by introducing biases in parameters when the assumptions are not met.
Therefore, for comparison a simpler approach of applying a triexponential decay was used to separate the originating ASL signal into compartments, which further provides compartmental transitional rate (k lin ).This approach was recently introduced as a proof of concept by Schidlowski et al, 23 where the authors applied a biexponential model to estimate k lin as a BBB permeability parameter in three subjects.The overall concept is based on compartmentalization of labeled water solely on the basis of change in transverse relaxation.This makes the approach more flexible where minimal assumptions are made without depending on any physiological parameters like CBF and transit times.
The PIB model has an ability to provide a voxelwise BBB permeability map which could be used in clinics to identify patterns of BBB integrity in various diseases.Since this ASL-based method uses the endogenous water in the blood as bolus, aided by water being a small molecule, it provides a possibility to detect minute and early changes in the BBB integrity.In this study, it was found that the whole brain BBB permeability in the older group was higher compared to the younger group.A pattern of relatively higher normalized difference of T ex between the groups was observed in the frontal areas of the upper brain slices.Cognitive decline associated with healthy aging has been linked to changes in the frontal lobe of the brain. 27Since no cognitive assessment was performed in this study, region-based analysis is needed to explore the coexistence of cognitive decline and faster water exchange in these brain regions.
GBCA DCE-MRI-based studies observed increased BBB permeability during normal aging. 28,29Moreover, in animal models increased BBB permeability to water due to age was recently reported using multi-TE ASL approach 9 and multi-flip-angle, multi-echo dynamic contrast-enhanced (MFAME-DCE) MRI method. 30In the latter study, authors compared water permeability with the GBCA DCE-MRI and it was suggested that the early changes in BBB permeability may be more sensitive to water as the DCE-MRI measurements were unable to detect them at the early stage of aging.Conversely, it should be noted that other variants of the ASL technique using DW-ASL have reported opposing trends of water exchange via the BBB due to aging.A recent study by Ford et al 12 reported a decrease in BBB water exchange rate K w ð Þ with age in humans.This contrast may be arising because of the variability in the intrinsic sensitivity of each  property being provoked to blood and tissue components or the differences in data acquisition method and modeling approach used.DW-ASL study did not consider intra-voxel transit time (ITT) which accounts for movement of labeled spins through smaller arteries and arterioles in the imaging voxel before exchange takes place.In this study, it was observed that the transit times, ATT and ITT, might change with age and such a difference in modeling approaches could have resulted in opposing trends.Moreover, Shao et al 13 directly compared K w measured using DW-ASL with GBCA  DCE-MRI results, and found significant correlation in only three brain regions which included white matter, caudate and middle cerebral artery, and not in the gray matter and rest of the five brain regions.The authors argued that the two techniques could be probing different underlying mechanisms for exchange of water and GBCA through the BBB.Multi-TE ASL could play an interesting role in unraveling the underlying mechanisms of water transport across the BBB, especially when comparing ASL methods where the tracer is the same, small water molecule.
The impact of aging on CBF and ATT has been extensively studied and the results of this study are consistent with previous aging studies of brain perfusion. 31,32Although, the mean CBF difference observed in this study, between the two groups, was higher than the reported range. 33One explanation could be that the macrovascular signal was not considered in this study, which may have affected the CBF values in the younger group.Moreover, same sampling times were used for the two groups, while a subject-specific adaptation of sampling times may provide a better estimation to resolve this issue.Furthermore, this study revealed that ITT was shorter in the older group relative to the younger group.ITT parameter accounts for transit through arteries and arterioles before labeled water reaches capillary exchange site.One reason could be the increased blood pressure 34 and increased narrowing of cerebral microvessels with age, 35 which may have contributed to increased blood velocity leading to reduced transit time in such vessel segments.
The simpler TD model also showed specificity for detecting BBB changes between the two groups.The older group showed a lower k lin value compared to the younger group.When comparing the group k lin values, we must consider where the difference in average k lin is coming from.This difference comes mainly from one TI, which is the earliest TI = 1600 msec.An important finding of this model is that the tissue fraction (f EV Þ at this TI was found to be significantly higher for the older group compared to the younger group.This could be interpreted as a quicker water exchange, already occurring at the earliest TI, resulting in a very high tissue fraction which did not change much in the later TIs.To explore this possibility further, a shorter bolus duration could provide a better estimation of bolus compartmentalization, but this could be limited to SNR considerations.Another explanation for this high f EV at TI = 1600 msec could be the narrowing of blood vessels or capillaries in the older group due to deposition of substances like cholesterol, in case of arteriosclerosis, or paramagnetic materials (e.g., iron), causing an apparent decrease in IV blood T2 which could further result in apparently high tissue fraction f EV .As discussed, both models showed sensitivity for detecting changes in BBB permeability with age.One important consideration is how realistically each model represents the underlying processes of water exchange and the interpretability of the resulting parameters.Drawing an agreement between the two approaches used in this study is not straightforward.Tissue fraction f EV at TI = 1600 msec and T ex showed a significant strong negative correlation, indicating that the volunteers who had high tissue fraction at the earliest TI, also showed lower T ex -supporting the interpretation of faster exchange via the BBB.On the other hand, k lin and T ex showed a strong positive correlation which shows opposing of the two models.Interpreting k lin is a bit complicated and the findings of this study suggest that it should not be taken alone as an exchange rate of labeled spins via the BBB.Lower k lin which apparently translates to a slower rate of water exchange in the older group, instead as shown by the data, is due to a quicker movement of labeled water spins into the tissue at the earliest TI.This mounts up to 70% of the total tissue fraction which changes very slowly (up to 85%) in the later TIs.k lin represents this rate of change for which the starting value (that is the amount of tissue fraction f EV at TI = 1600 msec) must be considered.If the starting value is relatively high, as in the case of older volunteers, the rate of change would be slow and the slope will be reduced, giving a low value of k lin .This shows that the simpler TD model, despite being able to detect age-based changes in the BBB between the two groups, may not be enough to fully understand the dynamics of the BBB due to its complex interpretation.This, on the other hand, highlights the importance of having physiological information included in the model which makes model interpretability global and more reliable.

Limitations
Low SNR is one of the known challenges of ASL imaging.Specifically, for the multi-TI approach, the tradeoff between scan time and SNR must be considered and, in this study, it was additionally affected by the bolus duration.Better SNR and higher ASL signal intensity could be achieved by using a longer bolus duration, which would increase TR and would ultimately result in a longer scan time.Moreover, in this study a version of Hadamard pCASL sequence was used which applied a fixed bolus duration for every TI.This also affected the increment between the TIs which was dependent on the bolus duration, unlike the conventional pCASL sequences.This limitation could be improved by adapting the Hadamard sequence to allow choosing different bolus durations to compensate for the T1 decay of labeled water as implemented by Schmid et al. 36 Furthermore, it was observed that the labeled blood arrived later in the older group compared to the younger group.This shows that the sampling times did not capture the optimal hemodynamics between the groups, such that many older volunteers had no inflow signal at TI = 600 msec.To fully utilize the potential of Hadamard imaging, a sequence with subject-specific optimization of sampling times can be used.One such method has been recently proposed, 37 where the sequence adapts to find optimal timings by analyzing intermediate images and adjusts bolus durations without the penalty of increase in the scan time.Such emerging methods offer a possibility to test subject-specific imaging which could ensure capturing meaningful information at every time point for every individual subject.
Both modeling approaches could be improved by using subject-specific T2 values to provide more reliable results.T2 values are reported to change with age. 38Increase in T2 relaxation with age has been linked with increase in free water content. 39Loss of neuronal cells and axons increase with age which facilitates the increased free water content in normal aging.Furthermore, blood oxygen saturation should also be considered when applying methods based on change in T2.Studies demonstrated high sensitivity of T 2 bl to the blood oxygen saturation 40 and the blood in capillaries is known to have less than 100% oxygen saturation. 11n this study, the scan time for the whole ASL data acquisition (HAD-8, HAD-4, and M0) was 8:14 minutes.This duration could further be shortened by replacing multi-TE HAD-8 protocol with a single-TE protocol.Multiple echo times are required only for exchange time estimation while HAD-8 is primarily designed to capture the inflow of the labeled blood for robust ATT estimation.This could reduce the total scan time to under 5 minutes (04:30 minutes) allowing the method to be tested and incorporated in a clinical setting.
Finally, there are various study design limitations.This study included only two groups with extreme age difference.A multi-center and multi-vendor larger cohort investigation, spanning over a continuous age range, is needed to better understand the phenomenon of age-related water exchange dynamics.Moreover, the study was conducted at 3T and a comparison at different field strengths is needed, as T2 relaxation difference between blood and tissue is reported to be affected with change in magnetic field strength.Furthermore, a longitudinal analysis with contrast-agent-based MRI and PET imaging is needed to validate the results.

Conclusion
The study showed age-related changes in BBB permeability derived from two different analysis approaches of multi-TE pCASL method.The increase in BBB permeability is in line with results reported in animal models investigated with a similar method as well as DCE-MRI.The ability of the multi-TE method to detect changes in BBB permeability suggests that the method could further be explored in pathologies affecting brain vasculature for testing its potential as an emerging BBB imaging biomarker.

FIGURE 1 :
FIGURE 1: Arterial spin labeling decoded images of a representative subject.The signal decreases over TIs mainly due to T1 relaxation and inflow of the labeled water while the decrease in signal over TEs occurs due to T2 relaxation.

FIGURE 2 :
FIGURE 2: Arterial spin labeling signal curve and its components generated using a multi-TE extended model for two different protocols used.The total signal is composed of signals originating from larger vessels-arteries and arterioles-where no exchange takes place for ITT duration and is denoted by "arteries signal."When the signal arrives at the capillary exchange site, it starts exchanging and the signal coming from blood in the capillary is called the "intravascular signal" and the signal which exchanges into the tissue is referred to as the "extravascular signal."(a) Signal curve representing HAD-8 protocol with SBD = 400 msec, ATT = 600 msec, T ex = 300 msec, ITT = 200 msec, and CBF = 60 mL/100 g/minute.(b) Signal curve for HAD-4 protocol generated using the same parameters except for a longer SBD = 1000 msec.HAD-8 protocol generated seven inflow times (TIs) while HAD-4 protocol resulted in three TIs, which are shown here for each representative younger (20 years old) and older volunteer (53 years old).

FIGURE 3 :
FIGURE 3: Group averaged parameter maps in MNI space.(a) BBB permeability maps in terms of exchange time T ex ð Þand T ex values compared between the two groups.(b) Shows cerebral perfusion (CBF) maps and mean values.(c) Arterial transit time (ATT) maps and mean values compared between the groups.(d) Shows intravoxel transit time (ITT) maps and mean values.All plots represent mean gray matter values and error in terms of standard deviation; where *P < 0.05 and **P < 0.01.

FIGURE 4 :
FIGURE 4: Group averaged exchange time T ex ð Þ maps along with the normalized difference map for different slices.Normalized difference appears to be high in the upper slices of the frontal area of the brain.

FIGURE 5 :
FIGURE 5: T2 fitted with (a and b) HAD-4 decoded arterial spin labeling (ASL) data and (c and d) Multi-TE M0 data for a representative subject.For ASL data voxels with RMSE >0.9 are included.(a and b) It can be observed that T2 of the labeled water decreases with increasing TIs.

FIGURE 6 :
FIGURE 6: Intravascular blood fraction (f IV ) and extravascular tissue fraction (f EV ) shown for increasing TIs of HAD-4.The intravascular fraction decreases over time as the TIs increase, whereas the extravascular fraction increases over the TIs.A threshold of RMSE >0.9 is used to include voxels showing reliable fit.Note that the TI = 3600 msec has a very low overall signal due to T1 relaxation of the labeled water.

FIGURE 7 :
FIGURE 7: (a) Extravascular tissue fraction (f EV Þ calculated for three TIs of HAD-4 (RMSE > 0.9).(b) Frequency histogram of f EV along with their peak frequencies detected at respective TIs.(c) The peak frequencies are fitted with a linear model to calculate slope k lin ð Þ representing the rate of change of tissue fraction over TIs as a measure of BBB permeability.

FIGURE 8 :
FIGURE 8: Comparison of slope k lin ð Þ values calculated with f EV tissue fraction.(a) Peak frequencies shown at three TIs for both groups.A trend of increasing frequencies can be clearly seen for the younger group.(b) k lin was significantly lower (0.054 AE 0.039 s À1 ) in the older group relative to the younger group (0.150 AE 0.038 s À1 ).(c) Group averaged slope of k lin calculated across subjects.Interestingly, the older group shows very high f EV of 0.7 at the earliest TI = 1600 msec which makes the overall slope significantly smaller (flatter) compared to the younger group slope; where **P < 0.01.

FIGURE 9 :
FIGURE 9: Correlation between the two model approaches.(a) Shows significant negative correlation (r = À0.80) between f EV at TI = 1600 msec and T ex , depicting that the volunteers who showed higher tissue fraction at the earliest TI, also showed faster water flux (lower T ex ) via the BBB.While, (b) and (c) show that the correlation between f EV at TI = 2600 msec and T ex ; and f EV at TI = 3600 msec and T ex was very weak and not significant (r = 0.18, P = 0.39 and r = 0.06, P = 0.78, respectively).(d) Shows significant positive correlation between k lin and T ex (r = 0.73).

TABLE 1 .
Summary of Mean Gray Matter Values of the Fitted Parameters Resulting From the PIB a Model Approach a Physiologically informed biophysical.b Exchange time.c Cerebral perfusion.d Arterial transit time.e Intravoxel transit time.