· Straightforward post-processing methods can lead to robust quantitative blood flow maps using ASL.

· Alternative post-processing methods can address potential problems and expand the capabilities of ASL.

ASL is based the difference in signal between label and control images, which is on the order of 1% of the gray matter signal. The intrinsically low SNR of ASL limits the achievable spatial resolution and imposes constraints on post-processing methods. Because the perfusion signal is the difference of two much larger signals, ASL is sensitive to bulk and physiological motion between label and control acquisitions. An important feature ASL is that it can produce quantitative perfusion maps, but there are a number of potential quantitation errors that must be minimized. Quantitative ASL is sensitive to variations in arterial transit time (ATT), which can lead to underestimation of perfusion. Quantitation requires a perfusion model with a number of unknown parameters, each of which must be estimated or measured. Despite these challenges, ASL is now a robust technique that is ready for broader clinical use. The ISMRM Perfusion Study Group and the European Consortium for ASL in Dementia recently published a consensus statement of the recommended implementation for ASL for clinical applications [1]. This article describes a practical baseline strategy that requires only simple post-processing and leads to perfusion images and quantitative perfusion parameter maps. In this talk, we will review the post-processing steps recommended in the consensus statement. We will then describe alternative post-processing steps designed to address potential problems and expand the capabilities of ASL. While many of the methods described were developed for ASL of the brain, they can typically be adapted for imaging in other regions of the body, such as kidneys or skeletal muscle.

The consensus statement recommends an ASL implementation that includes pseudo-continuous labeling, background suppression, and a segmented three-dimensional readout without vascular crushing gradients [1]. The post-processing steps required for this implementation are straightforward. The first step is to reconstruct the images using established methods for either 3D GRASE or 3D stack-of-spirals readouts. Then, perfusion difference images are generated using a simple subtraction of label and control images. The difference images are valuable clinically by revealing focal changes in perfusion, and the background suppression makes the subtraction less sensitive to motion. The next post-processing step is to generate quantitative cerebral blood flow (CBF) maps, which are valuable for detecting global changes in perfusion. The CBF maps are calculated using a simplified version of the standard kinetic model [2]. The simplifications of the model are based on several reasonable assumptions. One key assumption is that the entire labeled bolus is delivered to the target tissue, which is the case when the post-labeling delay (PLD) is greater than the ATT. While this assumption may be invalid in some patients, the authors recommend different acquisition parameters for different patient groups to minimize the probability of error. In this talk, we will review the simplified model and the recommendations for setting the parameters of the model. By following these recommendations, the result is a simple and practical quantitative perfusion imaging method.

There are many promising extensions and alternatives to the methods described in the consensus statement. One issue with 3D spin-echo-train readouts is that there can be substantial blurring in the partition-encoding direction, leading to partial-volume effects. This can be addressed in post-processing in two different ways. First, parallel image reconstruction can be used to shorten the echo train length, although at some cost in SNR. Second, blurring artifacts can be reduced using deblurring algorithms in post-processing [3]. In-plane blurring can be reduced by incorporating information from high-resolution anatomical images [4]. Accelerating readouts using a combination of parallel imaging and compressed sensing can enable single-shot 3D readouts, which can reduce scan time and motion sensitivity [5,6]. There has been substantial research into methods for perfusion quantitation, and we will give an overview of some of these during the talk. One example is the BASIL toolset [7], which employs a Bayesian approach to ASL quantification. Rather than acquiring images at a single PLD, dynamic ASL methods acquire images at a series of PLDs, which mitigates the possibility of underestimating perfusion because of long ATT and enables the computation of ATT maps. These methods require more acquisitions, but by using model-based compressed sensing methods it is possible to acquire each time point in 20 seconds [5].