Arterial spin labeling for the measurement of cerebral perfusion and angiography

Abstract

Arterial spin labeling (ASL) is an MRI technique that was first proposed a quarter of a century ago. It offers the prospect of non-invasive quantitative measurement of cerebral perfusion, making it potentially very useful for research and clinical studies, particularly where multiple longitudinal measurements are required. However, it has suffered from a number of challenges, including a relatively low signal-to-noise ratio, and a confusing number of sequence variants, thus hindering its clinical uptake. Recently, however, there has been a consensus adoption of an accepted acquisition and analysis framework for ASL, and thus a better penetration onto clinical MRI scanners. Here, we review the basic concepts in ASL and describe the current state-of-the-art acquisition and analysis approaches, and the versatility of the method to perform both quantitative cerebral perfusion measurement, along with quantitative cerebral angiographic measurement.

Keywords: Magnetic resonance imaging; angiography; arterial spin labeling; perfusion imaging.

Figure 1.

Part (a) shows the pulse sequence preparation elements for pseudo-continuous ASL (pCASL), consisting of a train of selective RF pulses (of low flip angle α and zero phase) in combination with suitable gradients in the direction of flow. The pulse train used as the control, which has pulses with the same flip angle but alternating phase, and which is substituted for the labeling RF pulse train, is also shown. Part (b) shows the repeating modulation of M_z as a function of the phase accumulated between RF pulses that is imposed on the flowing spins by the gradient shown in (a). Since the rate of accumulation of phase is proportional to the distance of the spins from the labeling plane, the x-axis is equivalent to distance. Note that since selective RF pulses are used, the flowing spins only experience the transition region shown in red, which causes their magnetization vectors to invert as they pass through the labeling plane (i.e. left to right in the figure). The dashed lines show the parts of the modulation pattern (locations) for which the spins are outside the influence of the selective RF pulses used.

Figure 2.

Schematic of some of the principal flavours of ASL pulse sequence. In each case, the green box shows the region of the label and the blue box shows the region of the control pulse. (a) shows a ‘conventional’ CASL pulse sequence and its accompanying label/control locations. (b) shows a PICORE PASL pulse sequence, in which the control pulse is an off-resonance pulse (not actually location specific). (c) shows an FAIR pulse sequence and labeling figure, in which the label is created with a non-slice-selective inversion pulse, and the control is a slice-selective inversion. Finally, (d) shows the QUIPSS-II PASL sequence, in which a saturation (red box) is added at time TI₁ to the labeling region to better control the temporal duration of the bolus of labeled blood.

Figure 3.

Multi-delay ASL strategies: in order to sample the dynamics of labeled blood flowing into the tissue to allow for arrival time estimation and kinetic model fitting, multiple label and control images can be acquired with different post-labeling delays (top). Alternatively, a Look-Locker strategy can be employed consisting of multiple low-flip angle readouts following each label or control preparation to sample the blood signal at different times after labeling (middle). In time-encoded CASL or PCASL, the labeling period is split up into a series of blocks. If these blocks are alternated between label and control states in different ways across different imaging cycles according to a Hadamard encoding matrix, then the perfusion signal arising from each of these separate labeling blocks can be ‘decoded’ in post-processing, yielding multiple images with different effective post-labeling delays (bottom). (Figure originally published in ‘Introduction to Perfusion Quantification using Arterial Spin Labeling’ by Michael Chappell, Bradley MacIntosh and Thomas Okell, Oxford University Press, and reproduced with permission).

Figure 4.

Schematic of vessel-selective ASL for labeling the four main brain-feeding arteries: vessel-selectivity can be achieved by targeting individual arteries of interest, using a labeling slab or spot, before moving on to the next artery to build up individual maps of the vascular territories by simple subtraction of label and control images (top). Alternatively, different combinations of vessels can be labeled or controlled across multiple imaging cycles, uniquely encoding the signal from each artery that can then be decoded in post-processing (bottom). Although more complex to process, the vessel-encoded approach results in perfusion signal from all arteries across all cycles, boosting SNR when the images are combined.

Figure 5.

Example time series plots from the ‘standard’ model for ASL kinetics comparing PASL and pCASL. Unless otherwise specified arterial transit time (Δt) = 0.7 seconds (PASL) or 0.9 s (pCASL), T₁ = 1.65 s, bolus duration = 0.8 s (PASL), 1.8 s (pCASL) following Alsop et al. The plots show arterial input functions with varying arterial transit time (a,b), residue function for varying T₁ (c), voxel magnetization for varying arterial transit time (d,e) and varying T₁ (f,g).

Figure 6.

Partial volume effects in ASL and methods for correction illustrated in simulated data created using PV estimates from structural data in an individual and different spatial distributions (a–c) of GM CBF (‘True’). ‘Standard’ refers to the estimated CBF map using standard quantifications, Spatial PV refers to the spatial method of Chappell et al. and LR corresponds to the linear regression method of Asllani et al. with different kernel sizes. Figure reproduced with permission from Chappell et al. (Figure 2).

Figure 7.

Example transverse maximum intensity projection from a 4D vessel-encoded ASL angiogram with bSSFP readout. Color represents the arterial origin of the blood signal: right/left internal carotid artery (red/green), right/left vertebral artery (blue/magenta).

Figure 8.

Schematic of strategies for dynamic ASL angiography: For PASL (top row), labeling (orange block) occurs quickly, so the addition of a time-resolved readout, in which the same set of k-space lines are repeatedly acquired (green blocks), allows the generation of images showing the inflow of labeled blood. Since only a subset of k-space lines can typically be acquired per ASL preparation, this process is repeated until all lines of k-space have been acquired for both label and control conditions. For CASL or pCASL, one method to obtain temporal resolution is to acquire a single image after labeling the blood, but to vary the label duration (middle row), which in turn varies the distance into the vascular tree that the labeled blood has travelled. Alternatively, a single long label duration can be used followed by a time-resolved readout. In this case, the first image shows the vascular tree filled with labeled blood and subsequent images show washout of the bolus. However, ‘inflow subtraction’, in which each subsequent image is subtracted from the first, results in images showing inflow rather than outflow for a more intuitive visualization. Images are shown in inverted contrast and control cycles are omitted for clarity.

Figure 9.

Preliminary data showing vessel-encoded pCASL perfusion imaging (top left) and angiography (bottom) in a patient with a history of transient ischaemic attack, alongside time-of-flight angiography data (top right) for comparison. Color represents the arterial origin of the blood signal as per Figure 7, with one example perfusion slice shown separately for each artery in grayscale for clarity. Reduced blood flow is apparent in the left vertebral artery (LVA, orange arrows). Significant collateral flow from the right internal carotid artery (RICA) through the posterior communicating artery to the right posterior cerebral artery is also observed (yellow arrows), and to a lesser extent on the left side (white arrows), perhaps to help compensate for the limited blood supply arising from the LVA. Adapted from Okell. Data were acquired in collaboration with Dr Ursula Schulz.

Figure 10.

Vessel-encoded pCASL dynamic angiography (top) and a time-of-flight (TOF) angiography maximum intensity projection (MIP, bottom) in a patient with arteriovenous malformation (AVM). Color represents the arterial origin of the blood signal as per Figure 7. Selected frames are shown demonstrating the major feeding vessels to the AVM from both the RICA via lateral (purple arrows) and medial (yellow arrows) branches of the right middle cerebral artery, and the LICA via the anterior cerebral arteries (orange arrows). The times shown are relative to the start of the VEPCASL labeling. Similar features can be observed in the TOF MIP, although some vessels are obscured due to overlying static tissue that is subtracted away in the VEPCASL data. Further work is required to separate the blood signals arising from smaller arterial branches, as has been demonstrated with super-selective ASL techniques, and to better visualize venous drainage. Adapted from Okell. Data were acquired in collaboration with Dr Natalie Voets.