This educational talk will cover common ASL post-processing steps. The talk includes (1) pre-processing of image data, (2) from general to basic models for perfusion quantification, (3) partial volume correction, and (4) advanced perfusion quantification. Focus will be put on the pre-processing and basic perfusion quantification.
Arterial spin labeling (ASL) exploits labeling of arterial blood water upstream to the tissue of interest, enabling noninvasive perfusion mapping. The labeled image is complemented with a control image with no labeling, and the difference between the two yields a perfusion weighted image. Hence, the most basic ASL post-processing is to perform this subtraction (controllabel) and then display the resulting difference image (ΔM). Since the difference signal is small, repeated label and control acquisitions and low spatial resolution is usually employed in ASL.
An attractive feature of ASL is that, using appropriate post-processing, quantitative perfusion maps can be calculated. This post-processing is however dependent on several factors, many of which constitute potential sources of error in the perfusion estimates, and care must thus be taken in how the data is acquired, how to perform pre-processing and what assumptions to make in the perfusion quantification.
The consensus of the ISMRM Perfusion Study Group and the European ASL in Dementia consortium on recommended implementation of clinical ASL (‘the ASL white paper’) [1] is a very good guide in this context since it summarizes the most common acquisition and post-processing methods.
Since ASL employ repeated acquisition of label and control images, motion correction is normally used. However, since the label and control images have slightly different contrast, the motion correction approach may not straight-forward.
Although the ASL white paper recommends 3D readout, some researchers still use 2D readout. For 2D readout, slice time correction is required due to the different readout times of the different slices. Strictly speaking, slice time correction should be performed prior to motion correction.
Many of the readout modules used in ASL gives rise to geometrical distortions, and it may be of importance to correct for them. One approach, as often employed in fMRI with EPI readout, is to map the B0 field and use a forward model of the distortion process to perform the correction. Another approach, which is easier to implement on the scanner, is to acquire an additional reference (M0) image with opposite phase encoding direction (i.e, so that 2 M0 images with PA&AP or LR&RL are available). The distortion field can then be inferred and used to correct all ASL images in the data set.
Outlier removal and denoising have often been used in ASL in an attempt to boost the effective SNR. Outlier removal often works by identifying image volumes, slices or individual voxels (in the repeated data set) that seem to contribute more to uncertainty than to improved SNR. The most basic approach is to manually or automatically remove ΔM images that are unlike the mean ΔM image. Denoising can be achieved with many different algorithms, and the goal is to remove noise (thermal and/or physiological) from the raw data (control and label images).
Originally, ASL was implemented as a steady-state experiment, and in this context, perfusion contrast was identified and modeled as a change in the apparent T1 of tissue [2]. Later, ASL analysis shifted towards a more bolus experiment oriented approach, and Buxton et al. generalized this by describing ASL from the perspective of tracer kinetic modeling, which is summarized in the so-called general kinetic model for ASL [3]. This general model can then be used to derive perfusion models for specific ASL techniques and using certain assumptions. The standard model is a common set of assumptions leading to tractable ASL signal equations. ASL is usually implemented as a single time-point experiment and using some additional assumptions the basic (single time-point) ASL model is obtained [1].
The magnetization difference ΔM constitutes our tracer concentration (ASL signal), and thus, the first step of the perfusion quantification is to calculate it. Simple subtraction is the most common approach. Secondly, acquisition parameters and assumed fixed parameters are entered into the model. Third, the perfusion is calculated using the appropriate model. To yield quantitative values, a reference (M0) image is included in this last step.
Partial volume correction (PVC) is an extra post-processing step that has gained interest in the ASL community recently. Since ASL employs low spatial resolution, the actual tissue perfusion (in mass-specific units, i.e. ml blood/100 g tissue/min) in a voxel is convoluted with the amount of tissue in that voxel. For example, for brain perfusion, partial volume effects (PVEs) correspond to voxels with mixture of cerebrospinal fluid (CSF), gray matter (GM) and white matter (WM). By estimating partial volume (PV) maps, the perfusion maps can be corrected for PVEs using a mixed perfusion model solved with a PVC algorithm. There are a few PVC algorithms available of variable complexity.
The PVC thus separates the effect of local tissue volume and tissue perfusion. This separation can be of considerable importance, especially for conditions in which a volumetric tissue alteration is plausible, for example, cerebral atrophy in the elderly or in connection with neurodegenerative diseases. Furthermore, PVC could potentially reduce the inter- and intra-subject variation of perfusion values since it reduces variability originating from tissue volume and PVEs.
To reduce the sensitivity to inter- and intra-subject variation in arterial transit time (i.e., the time it takes for the labeled blood to reach the tissue of interest), multiple time-point (multi-TI) ASL can be used. In this case, a kinetic model is fitted to the measured ΔM curve, and estimates of both perfusion and arterial transit time can be obtained. This type of analysis can also include other effects such as non-ideal label duration and arterial dispersion.
Conventional ASL techniques can be categorized into pulsed ASL (PASL), continuous ASL (CASL), and pseudo-continuous ASL (PCASL). Normally, techniques belonging to one of those categories can be analyzed with the post-processing approaches discussed above. Other special ASL techniques such as model-free ASL, velocity-selective ASL (VSASL), acceleration-selective ASL (AccASL), vessel-encoded ASL (VEASL), and dynamic ASL (DASL) require separate post-processing.
[1] Alsop DC et al. Recommended implementation of arterial spin-labeled perfusion MRI for clinical applications: A consensus of the ISMRM perfusion study group and the European consortium for ASL in dementia. Magn Reson Med 2015;73:102–116
[2] Williams DS et al. Magnetic resonance imaging of perfusion using spin inversion of arterial water. Proc Natl Acad Sci USA 1992;89:212–216
[3] Buxton RB et al. A general kinetic model for quantitative perfusion imaging with arterial spin labeling. Magn Reson Med 1998;40:383–396