Introduction

Velocity Selective ASL^1,2 is insensitive to transit delay (TD) compared to conventional ASL techniques. However, its relatively low labeling efficiency due to saturation instead of inversion has limited its use. Significant efforts that have been made to overcome this problem include the implementation of multiple velocity selective saturation (VSS) modules (mm-VSASL)² and Fourier-transform based velocity-selective inversion (FT-VSI)³. Both methods report significant SNR improvement over original VSASL method using single VSS module. The purpose of this study is to investigate the performance of mm-VSASL and FT-VSI under various off-resonance (B₀) and B₁ conditions using Bloch simulation. A few variants of VSI pulses were also compared, including a new modified pulse.

Methods

Definition of cutoff velocity
In order to have a meaningful definition of cutoff velocity that works for both the saturation- and inversion-based labeling methods, the ASL signal (label/control difference) was calculated and then convolved with a laminar flow pattern⁴ to give a Sig_ASL-V_mean profile. Then following the definition in Wong et al¹, the cutoff velocity was defined at the first crossing point where Sig_ASL = 1 for both VSS and VSI modules (Figure 1a). The gradient amplitude for all the sequences were adjusted to have a cutoff velocity of 2 cm/s.
Weighting by a realistic arterial velocity/volume distribution
The Sig_ASL-V_mean profile was then weighted with an arterial blood volume distribution measured in human brain over a big velocity range⁷ (Figure 1b) to calculate the total ASL signal and the signal per unit of time (Sig_ASL,tot/√(2TR), or SNR efficiency). This was done for different values of B₀ (from -200 Hz to 200 Hz with an interval of 50 Hz) and B₁ (from 70% to 130% of the nominal B₁ with an interval of 10%). The obtained results were then weighted by a 2D gaussian distribution (σ_B0 = 200 Hz, σ_B1 = 0.3, µ_B0 = 0, µ_B1 = 1, ρ = 0) to calculate the mean signal and signal per unit time. The standard deviation (SD) was calculated without weighting.
Bloch simulation with B₀ and B₁ variation
Single module velocity-selective saturation (VSS) with DRHS pulse¹ and sym-BIR8 (VSS-sBIR8)⁵, two-module VSASL (with sym-BIR8, VSS-sBIR8 x2)², original VSI sequence (VSI-orig)³, a modified VSI pulse using maximum-phase sinc modulation (VSI-mp-sinc)⁶ and a new modified VSI pulse using regular sinc modulation (VSI-sinc, Figure 2) were investigated. The performance, i.e. SNR efficiency, was simulated using Bloch simulations in MATLAB2019b (The Mathworks, Nantick, MA). The 1xVSS-sBIR8⁵ pulse with nominal B₁ scaling = 1 and off-resonance = 0 Hz was considered as the reference for comparison. The other parameters used in simulation included: T₁ = 1660 ms, T₂ = 150 ms, TR = 4500 ms, TI (single module) = 1300 ms, TI₁/TI₂ (two modules) = 1150/820 ms². Additionally, for the two-module case, TI₁ + TI₂ < BD was considered². The magnetization response with respect to velocity (Sig_ASL-V_mean) for VSS-sBIR8 and the VSI pulses were simulated for both label and control conditions and subtracted to yield ASL signal.

Results

Sig_ASL-V_mean responses shown in Figure 3 demonstrate that the modified VSI pulses contributes to better Sig_ASL-V_mean profile than the original VSI pulse. The overall SNR efficiency and their SDs for all the pulses simulated are listed in the table in figure 4.
The two-module VSS and two modified VSI pulses provided over 30% SNR efficiency increase over the single-module VSS. The modified FT-VSI pulses showed higher overall SNR efficiencies compared to the original FT-VSI pulse, with VSI-sinc performing the best with a 35% increase. VSI-orig pulse only showed a 20% increase compared to the single module VSS-sBIR8 pulse. When compared with the single-module DRHS pulse, the increase of 28.9% was consistent with the results reported previously³.
The weighted mean SNR efficiencies and SDs along off-resonance and B₁ were calculated to further investigate the dependence of the performance on these variations (Figure 5). The two-module VSS-sBIR8 was the most robust over the ranges of B₀ and B₁ simulated as the mean SNR efficiency doesn’t change significantly, and the SD values were much lower (~ 1/10th) compared to all the VSI pulses (Figure 4). The modified FT-VSI pulses showed better SNR than VSS-sBIR8 only when B₁ variation was relatively small (0.9 ~ 1.2). VSI-orig showed vulnerability to both off-resonance and B₁ variation, whereas the modified VSI pulses seemed to have reasonable stability along off-resonance, while both being sensitive to B₁ variation. VSI-sinc showed the best stability among the three VSI pulses.

Discussion & Conclusion

Improvement of the modified VSI pulses is likely due to the use of composite refocusing pulses, they also have less SNR efficiency reduction when B₁ is higher than lower, as also observed previously⁶. As all the VSI pulses under investigation are B₁ sensitive, it is worth exploring strategies to improve their B₁ insensitivity. Alternatively, mm-VSS using sym-BIR8 provides a significant amount of increase in SNR efficiency while being the most robust across a big range of off-resonance and B₁ variation. It should also provide a more accurate quantification than FT-VSI pulses.