The VSASL signal arises from spins with |V| < Vc, and the choice of Vc is therefore a key factor in the localization of the VSASL signal. In the ideal case, Vc would be chosen to correspond to the velocity of the arterioles that feed into capillary beds. This would restrict the VSASL signal to capillaries and tissue, thereby guaranteeing excellent localization of the signal to the target tissue. But the velocity of blood entering capillary beds is approximately 1 mm/s, and the use of a cutoff velocity that is this low would result in severe sensitivity to subject motion, brain pulsatility, and diffusion effects (see later) and would not be useful. In practice, a cutoff velocity of approximately 1 cm/s is achievable without severe sensitivity to these sources of artifacts. For most of our measurement, Vc = 2 cm/s. The vessel size that corresponds to this velocity is approximately 50 μm,⁶ and in the brain this is the approximate size of the arteries that penetrate the cortical surface. Thus, for this value of Vc, the VSASL signal will be localized downstream from these penetrating arteries and therefore appear in brain tissue, rather than in surface arteries or farther up the arterial tree. Use
of higher Vc results in further reduced sensitivity to artifacts but also in signals appearing farther upstream. As in all ASL methods, the VSASL signal is not guaranteed to appear in the voxel in which the perfused tissue resides but is more likely to do so for low values of Vc. In some implementations of conventional pulsed or continuous ASL, flow-weighting gradients are added in the imaging pulse sequence specifically to reduce artifacts from upstream arterial blood and can be identical in implementation to the second velocity-selective pulse in VSASL. Although in VSASL it is an integral part of the scheme to quantify CBF by performing |V| < Vc filtering, however, in conventional ASL it can complicate quantitation by saturating an unknown amount of upstream tagged blood.

In some pathologic states, slow-flow conditions may exist, which place the spins with |V| = Vc further upstream than in healthy subjects, resulting in a displacement of the VSASL signal upstream. For this reason, the optimal choice of Vc may depend on the range of velocities present in a particular application.

Because the velocity cutoff is the same for both the tagging process and the velocity filter that is applied prior to image acquisition, there is no gap, either in velocity space or physical space, across which the tagged blood must travel to produce a VSASL signal. This is the most important distinction between VSASL and conventional pulsed or continuous ASL, where there is a spatial gap between the tagging slab or plane and the imaging volume, resulting in a transit delay. Indeed, the arteries that are tagged in VSASL are often within the imaging volume itself. The VSASL signal is thus inherently insensitive to transit delays and, in principle, capable of accurate measurement of CBF in slow and/or circuitous routes of circulation in disease. In conventional, spatially selective ASL, slow flow results in increased transit delays. These delays require increased PLDs to avoid quantification errors and result in lower SNRs owing to label decay T₁ relaxation during the PLD time, as well as the potential for residual errors when the maximum potential transit delays are underestimated. In VSASL slow flow does not result in an artifactual decrease in the overall ASL signal, but in an upstream displacement of the detected signal as described in the previous section on localization.

In VSASL, velocity-selective pulse trains are used to saturate flowing spins. The design of these pulse trains is discussed in more detail later, but the general idea is that flow-weighting magnetic field gradients (similar to phase-contrast or diffusion-weighted MRI) are used to modulate the phase of flowing spins, and that a nonuniform distribution of velocities due to laminar, turbulent, or multidirectional flow produces phase cancellation and results in saturation of the net magnetization among flowing spins. Phase accrual will only occur if spins are moving along the direction the flow-weighting gradient is played out. Thus, using this approach, velocity-selective saturation is sensitive to only one direction of flow, defined by the direction of the velocity-encoding gradients. Unlike diffusion-weighted imaging, however—where usually three or more measurements with different diffusion-encoding directions are used—it is not necessary to repeat the VSASL measurement in different directions to correctly measure perfusion. As described previously, the choice of Vc determines the location in the vasculature at which spins upstream from this location are tagged by saturation and spins downstream are imaged. This location also depends on the direction of velocity encoding, because it is the location at which the component of the flow velocity in the direction of the velocity encoding decelerates through Vc (see Fig. 20.4). For vessels that are aligned with the direction of the velocity-encoding gradient, this will occur farther downstream, whereas for vessels that are nearly perpendicular to the velocity-encoding gradient, this will occur farther upstream. Empirically, in healthy subjects, the VSASL signal is relatively independent of the direction of velocity encoding for Vc < 4 cm/s but is significant for higher Vc.² The choice of Vc necessary to achieve direction independence in slower flow will probably be lower, and the optimization of this parameter will have
to be determined empirically for applications in which slow flow is likely.

In VSASL, the difference between tag and control conditions is that in the tag condition, flow-weighting gradients are applied in the velocity-selective tagging pulse module, whereas in the control condition gradients are not applied. This produces the desired saturation of flowing spins in the tag condition but also imparts a small amount of diffusion-related attenuation.^2,3,7 For Vc = 2 cm/s, the b-factor of this diffusion weighting is approximately 1 s/mm², which results in a signal attenuation in gray matter of approximately 0.1%. Although this attenuation is relatively small, it is not negligible compared to the VSASL signal itself, which is approximately 1%. Moreover, diffusion attenuation in cerebrospinal fluid (CSF) is approximately three times higher than in gray matter, and because CSF is in close proximity to gray matter, particularly over the cortical surface, diffusion-related tag–control signal differences in CSF can contaminate the VSASL signal. Several approaches have been taken to minimize or correct for this effect. First, for a given flow weighting, the diffusion b-factor is reduced with a longer time separation between the two lobes of the flow-weighting gradients.² This is because the flow moment varies linearly with both time separation and gradient area, whereas diffusion weighting varies linearly with the time separation but quadratically with the gradient area. Thus, for a given flow moment a longer time separation with a smaller gradient area results in lower diffusion weighting. The tradeoff is that a longer time separation also results in increased T₂ decay, which directly reduces the VSASL signal. In addition, pulse sequence modifications have been introduced either to null the CSF signal by inversion⁸ or to measure and remove the contribution of CSF to the VSASL signal.⁹