Arterial spin labeling (ASL) MRI is based upon inversion (or saturation) of the magnetization of arterial blood in the neck region (see Figure 1)¹. After a delay-time of approximately 1.5-2 seconds, to allow the blood to travel to the brain tissue, a proton-density weighted scan is made: the label image. Subsequently the same procedure is repeated, but without inverting the inflowing blood: the control image. Subtraction of the label from the control image will cancel out signal from static tissue, leaving only the signal from the labeled spins, i.e. the blood that has flown into the brain tissue.Traditionally, two different types of ASL are recognized: pulsed ASL that labels a large region below the imaging slices with a short (≈15 ms) inversion module and (pseudo-)continuous ASL that labels for a longer time (≈2 s) all blood that flows through an inversion plane. In the consensus paper of 2015, pseudo-continuous ASL (pCASL) was designated as workhorse technique for ASL imaging^4,5. Main reasons for favoring this implementation of ASL are the easy implementation on clinical MR scanners, easy quantification because of the well-defined temporal width of the labeling bolus, the higher SNR due to the continuous approach, and the easy post-processing/interpretation of the data. Throughout this abstract and presentation, pCASL will be the assumed method except when explicitly stated otherwise. Relatively recently, a third category of ASL (spatially non-selective (SNS) ASL) was proposed with velocity selective ASL being the most-applied example^6,7. In SNS-ASL, labeling is not performed below the imaging slice, but based on the flow properties of blood by means of motion-sensitizing gradients. This results also in creation of label within the imaging volume and thus much closer to the tissue, thereby shortening the time for the label to travel towards their end destination in tissue.

Important for understanding the quantification of ASL is the fact that the non-invasively created tracer (the labeled blood) are in fact the water-molecules in blood, which both quickly exchange magnetization between blood plasma and blood cells, as well as between the intra- and extravascular compartment in brain tissue at the capillary level^2,3. The tracer employed in ASL is therefore not only an endogenous, but also an (almost) freely diffusible tracer (i.e. when arriving in the microvasculature the magnetization will almost immediately exchange with the extravascular compartment). This provides the basis for the quantification approach: after creating a bolus of labeled spins with a pre-determined temporal width, the label will be transported to the supplied brain tissue, extravasate into the tissue and accumulate (for the largest part) in the brain tissue until it will be detected, i.e. the local cerebral blood flow is proportional to the amount of label accumulated in a voxel. Three main processes need to be taken into account when quantifying the blood flow:

1. The efficiency of labeling: pCASL does provide on average a high labeling efficiency, but the labeling efficiency can be affected by the velocity of the blood and B₀-inhomogeneities resulting in low or even negative signal on the CBF-maps. Normally, a constant labeling efficiency is taken into account as obtained from simulation studies of the ideal situation. This will of course not correct for artefactual low labeling efficiency, due to for example off-resonance effects. Solutions that have been proposed include the measurement of the blood velocity by quantitative phase contrast MRI and using this as input to the simulations⁸, by equating the total cerebral blood flow to a reference flow measurement⁹, or by direct measurement of the labeling efficiency in a separate scan¹⁰. When no further corrections are employed, it is at least essential to realize when interpreting pCASL-scans that sub-optimal labeling might have occurred.

2. Decay of the label: Labeling is performed by inversion of the magnetization and therefore label will be lost via longitudinal relaxation processes. Because of the relatively long time between labeling and detection of the label in the brain tissue (typically longer than 2 seconds), one needs to correct the detected signal for the T₁-relaxation of the label during this delay-time. Since the label will be located for the longest time in arterial blood, the consensus paper proposed to assume that the label was exclusively located in blood and therefore only to correct with the T₁ of blood (approximately 1.65 s at 3 Tesla). However, several factors can influence the T₁ of blood with main magnetic field strength, hematocrit and oxygenation as the three most important factors^11,12. Hematocrit is known to be different between males and females and in several diseases changes in hematocrit have been reported. When observing differences in quantitative ASL-values between two groups of subjects, it is therefore important to consider whether differences in hematocrit (and thus T₁ of blood) might explain the observed differences or that a difference in CBF is a more likely explanation. Secondly, in most subjects the label will at some point in time (approximately 1.5 – 2.0 s after labeling) move from the intra- to the extravascular compartment and from that time onward the label will relax quicker, i.e. with the T₁ of tissue (approximately 1.2 s). Differences in the time that it takes for the label to travel through the vascular tree and to exchange magnetization with the extravascular compartment, might therefore also introduce differences in observed ASL-signal.

3. Normalization of ASL-signal: MRI is traditionally a qualitative imaging modality, and the obtained signal after subtracting label from control images, needs therefore to be calibrated. In principle, calibration should be performed to the magnetization of a voxel containing pure arterial blood, i.e. the maximum ASL-signal that could be observed. In practice such a voxel is not available and it has therefore been proposed to use a minimum contrast reference scan obtained with the same readout parameters as the ASL scan, but a TR of 2-4 s⁴. The use of a reference image instead of a signal value, opens-up the possibility to perform a voxel-wise normalization of the signal-intensity that will also correct for most signal-variations due to local differences in T₂^(*) or coil sensitivities.

Two main artefacts should be expected when working with ASL¹³. The first is the occurrence of motion artefacts, which do result in strong artefacts because ASL is a subtraction technique. Motion correction in post-processing can alleviate some of the artefacts, but typically one would still see some ringing around the brain when the subject moved a lot. Optical tracking and prospective motion correction techniques have been proposed as solutions to minimize motion sensitivity. Moreover, probably an equally important solution is the use of background suppression in the ASL-sequence, which was also included in the consensus statement as part of the suggested standard implementation of ASL. By lowering the signal intensity in the brain tissue, the effect of a mismatch in static brain tissue after motion is lowered, thereby improving image quality.

The second important artefact is known under the names “transit time artefacts” or “vascular artefacts” (see Figure 2). These artefacts occur in subjects with delayed blood flow, resulting in the presence of label in the major arteries during readout of the signal instead of the label being present in the microvasculature or brain tissue. This is a violation of the assumption that all labeled blood has arrived in the brain tissue and will lead to an underestimation of the tissue-perfusion and hyperintense signal within the arteries. One solution to avoid such artefacts is to prolong the delay-time between the end of labeling and the readout (i.e. increased post-labeling delay time (PLD)), but this will automatically decrease the SNR of the resulting ASL-image, since the label will have had more time to relax than for a shorter PLD. Whereas one could also use flow-crushers to get rid of the signal in the large vessels, this would also obscure information, i.e. the fact that label could have been observed in the large arteries, tells us that label was still on its way to feed the brain tissue and tells us to take this into account in our interpretation of the ASL scan. One should therefore refrain of interpreting an ASL-scan just as an perfusion scan, but keep reminding oneself that one is looking at an “ASL-scan” of which transit time artefacts are an important indicator of impaired large vessel transport.

The audience will be taken in an interactive way through the acquisition, post-processing and interpretation of ASL scans. Many examples will be discussed.