Introduction

Blood-brain barrier (BBB) dysfunction is increasingly recognised in a number of neurological conditions, including stroke, multiple sclerosis and neurodegeneration^1-3. Estimates of water exchange across the BBB have demonstrated greater sensitivity to subtle damage than conventional approaches that use leakage of gadolinium-based contrast agents (GBCA)⁴ , thus presenting the possibility of detecting disease earlier.

Contrast-enhanced arterial spin labelling (CE-ASL) has recently been proposed⁵ as a method for quantifying the permeability surface area product of the BBB to water (PS). Acquiring ASL images in the presence of an intravascular contrast agent should in theory enable the label location, in either the intra- or extravascular space, to be determined as a function of post labelling delay (PLD) time due to the difference in blood and tissue T₁. However, the shorter blood water T₁ relaxation time causes the difference signal to decay more quickly, leading to an inherent trade-off between GBCA dose and signal-to-noise ratio (SNR). In this work, we: (i) investigate the optimal T₁ of blood post-contrast for quantifying BBB water exchange; (ii) use simulations to assess the accuracy and precision of PS at various SNRs, and; (iii) demonstrate the optimised method in a healthy volunteer.

Methods

Simulations
Signals were generated using a two-compartment model accounting for finite water exchange between intra- and extravascular compartments⁶ (see Fig. 1). Optimal sensitivity to water exchange was explored by varying the relaxation time of blood post-contrast between 0.1<$$$\text{T}_{\text{1,b}}^{\text{post}}$$$[s]<1.7. Other parameters were^1,7: relaxation time of blood pre-contrast $$$\text{T}_{\text{1,b}}^{\text{pre}}$$$=1.7s; relaxation time in extravascular space T_1,e=1.5s; blood flow f=60ml blood(min)^-1(100ml tissue)^-1; PS=6.6ml water(min)^-1(ml tissue)^-1; blood volume v_bw=0.05ml blood(ml tissue)^-1; arrival time t_A=1.2s. Fitting was performed on 2500 noisy signals using a least-squares minimisation; T₁ times were fixed to their ground truth values, leaving as free parameters PS/v_bw, f and t_A. The GBCA dose needed to achieve optimal $$$\text{T}_{\text{1,b}}^{\text{post}}$$$ was explored as a function of time post-injection, and calculated using the mean vascular input function from the saggital sinus of 32 healthy volunteers (mean age 66 years, range 52-81 years) following 0.1mmol/kg injection of Dotarem GBCA.

Data acquisition
Proof-of-concept data were collected in a single volunteer (64 years, female) on a simultaneous PET-MR scanner (GE Healthcare) using Enhanced ASL with pCASL labelling, 3D spiral FSE, 6 PLDs and variable label duration (LD) (Fig. 1), with voxel size 1.7mmx1.7mmx4mm. A 0.1mmol/kg dose of Dotarem GBCA was injected and the ASL acquisition repeated (3 PLDs; Fig. 1) ~20min later. T₁ mapping was performed (with B₁⁺ correction) pre- and post-injection using a 3D T₁-weighted spoiled gradient echo variable flip angle approach, with flip angles 2, 20, 30 and 40 degrees and TR/TE = 13.5/1.08ms. T_1,b was estimated from the superior sagittal sinus. A 1mm isotropic T₁-weighted image was acquired to segment grey matter, enabling the mean ASL signal to be extracted for each PLD. The two-compartment model⁶ was fitted to the data to estimate PS/v_bw, f and t_A.

Results

Fig. 2a shows that PS/v_bw was estimated with the greatest accuracy and precision for 0.7<$$$\text{T}_{\text{1,b}}^{\text{post}}$$$<1.0s; this was particularly evident at low SNR. Minimal bias in PS/v_bw was observed by varying T_1,e (Fig. 2b).

Fig. 3 demonstrates the expected region of interest (ROI) scale at which PS/v_bw may be estimated with reasonable accuracy and precision. Using this experimental approach clinically (that is, assuming SNR~25 in the control images), global estimates of PS/v_bw across gray matter (GM) could be expected with a CoV~45%. A 4-fold increase in SNR would allow PS/v_bw to be quantified in individual cortical regions with a similar level of precision.

Fig. 4 shows how the optimum T_1,b may be obtained as a function of GBCA dose and time after injection.

In vivo images are shown in Fig. 5. Images post-contrast show the expected reduction in the ASL subtraction signal, but parenchymal structure remains visible. Averaged across all GM voxels, parameter fits were: f=33.6mL/min/100mL, t_A=1.36s and PS/v_bw=45min^-1.

Discussion

This work evaluates the feasibility of water exchange measurements using CE-ASL. We explored the required reduction in $$$\text{T}_{\text{1,b}}^{\text{post}}$$$ for optimal measurements of PS/v_bw, and demonstrated that the method is robust across a range of values. Given the inherently low SNR of ASL-based techniques, it was important to assess the magnitude of measurement errors in order to evaluate experimental design limits. Simulations indicated that global estimates of PS/v_bw should be achievable clinically with reasonable accuracy and precision; this finding was corroborated in vivo, where global GM PS/v_bw estimates in a healthy volunteer were in line with published values¹.

A practical application of this method would be its use in conjunction with DCE-MRI, particularly when BBB damage is minimal. Following a DCE-MRI acquisition lasting at least 20min, $$$\text{T}_{\text{1,b}}^{\text{post}}$$$ would have recovered enough to generate suitable contrast for subsequently estimating PS/v_bw using CE-ASL.

Conclusions

Measurements of water exchange across the BBB are important for characterising subtle damage early during disease onset. We present a CE-ASL method for clinically-feasible global GM measurements of PS/v_bw in 13min. Simulations indicated that maximal accuracy and precision could be obtained across a range of $$$\text{T}_{\text{1,b}}^{\text{post}}$$$ values, thus providing flexibility in experimental set up. By utilising residual T₁ relaxation rate changes from conventional DCE-MRI protocols, CE-ASL measurements could offer an efficient approach for assessing BBB permeability.

Acknowledgements

Thanks to GE Healthcare for use of the Enhanced ASL sequence. This work was supported by EPSRC grants EP/S031510/1 and EP/M005909/1.