A neuroradiologist’s guide to arterial spin labeling MRI in clinical practice

ASL’s aim is the assessment of tissue perfusion rate, which is very different from macrovascular blood flow. Tissue perfusion, or the exchange of water and nutrients with the tissue, happens along the entire length of the capillaries (Fig. 1). ASL basically ‘follows’ blood water molecules from the arterial compartment all the way to the tissue capillary bed, using them as a free diffusible tracer. ASL is easily carried out by the inversion or saturation of the magnetisation along the Z-axis of blood water molecules in the feeding arteries. This part of the ASL acquisition is called the labeling. Following the labeling, time is allowed for the blood to travel to the tissue: the so-called post-labeling delay (PLD) or inversion time for certain ASL techniques. The delay is chosen such that images are ideally acquired at the time of exchange of the water molecules with the tissue magnetisation (Fig. 2). Arterial blood labeling is achieved by a combination of RF pulses and gradients in order to invert the longitudinal magnetisation (T1) of blood water protons.

After the PLD, the image acquisition phase starts to obtain signal from those labeled protons coming from the feeding arteries into the tissue. Different types of readout can be employed. In order to differentiate the labeled magnetisation signal from the static tissue magnetisation signal, two images are acquired: one with (labeled image, Fig. 2a) and one without (control image, Fig. 2b) labeling. Subtraction of the labeled from the control images provides the perfusion-weighted image.

Most ASL techniques aim to avoid signal coming either from the flow in vessels or from static tissue, employing for that purpose additional gradient and RF pulses. In addition, numerous pulse sequences have been developed to maximise SNR by reducing all confounding and potentially artefactual signals. In particular, multiple cycles of control and labeled images are acquired, leading to typically 4–6-min acquisition times. The need to acquire multiple repetitions may however lead to additional artefacts, such as those due to patient motion, and necessitate the use of additional strategies to obtain usable ASL data either during acquisition or with post-processing techniques.

Several physical and physiological parameters will affect the quality of the ASL image and need to be understood prior to choosing the most appropriate sequence and parameters. Relevant factors include the labeling efficiency, arterial blood T1 and relaxation times, the blood transport time through vessels and tissue (depending on flow velocity), and finally the so-called magnetisation transfer effects.

Arterial labeling efficiency is the first key aspect to a good ASL acquisition. How effectively we invert the longitudinal magnetisation directly affects the eventually acquired signal. Efficiencies, designated with α, can range from 80 to 98 % with currently available pulse sequences, which need to be taken into account when quantifying CBF [2, 17‐20].

Both blood transit time and T1 relaxation are in the order of seconds, meaning that about two thirds of the label will have decayed by the time the blood reaches the capillary bed. The optimal PLD is thus almost invariably a compromise between the arterial blood T1 and transit time. Arterial T1 and feeding arteries transit times both need to be taken into account to get a proper absolute quantification by physiological modelling [21]. Ideally, the arterial blood relaxation time (T1_Blood) is measured on an individual basis. In the capillary bed, the arterial blood water will exchange with the surrounding tissue. The ratio of tissue water versus intravascular water in the brain is in the order of 20:1, meaning that only a very small proportion of the measured water signal is affected by the magnetic label. When the labeled water enters the tissue, over 90–95 % exchanges with the tissue water instantaneously. For this reason, it is impossible to measure any labeled venous blood flowing out of the tissue, due to the very short T1 relaxation time compared with the equilibrium time (1–2 s versus 10–15 min for a nominal CBF of 50 ml/min/100 g, respectively) [22].

One of the major confounds when using ASL comes from magnetisation transfer effects. The application of off-resonance RF pulses to label the arterial blood proximal to the tissue of interest will induce a reduction of MR signal that mimics the signal decrease induced by the labeled blood water exchange. This is due to the fact that protons in macromolecules have a very short T2 and thus a very broad frequency spectrum. As such, the off-resonance RF pulses used in any asymmetrical ASL sequence will inadvertently not only saturate the arterial blood but also protons in these macromolecules. This saturation will then be transferred to tissue water molecules reducing the MR signal [23].

Three main categories of ASL pulse sequences have been developed, each aiming to take these factors into account with variable success: continuous ASL (CASL), pseudocontinuous ASL (pCASL), and pulsed ASL (PASL). PCASL is at present the sequence of choice, due to its high labeling efficiency combined with its ease of implementation and hardware specifications for clinical scanners [16].

The white paper recommends a pCASL labeling scheme, combined with background suppression and a segmented 3D readout. SNR will be better at high field, so 3 T is preferred, especially in patient populations with slow or poor blood flow [16]. At 1.5 T, ASL is feasible when blood flow is not too compromised and in the paediatric population but will require longer scanning times to obtain sufficient signal.

Phased array coils (eight or more channels) are recommended to provide higher SNR especially in the cortex. These coils also allow parallel imaging techniques and may thereby accelerate the acquisition [33].

The measurement of perfusion is probably most directly applicable to cerebrovascular disease and ischaemic episodes. These conditions typically manifest themselves on ASL as regions of altered perfusion or delayed arterial transit time (ATT) because malformed, occluded, and damaged vessels perfuse less labeled water than normal vessels [41]. ASL has been explored for the non-invasive detection of cerebrovascular abnormalities, identification of ischemic events, and risk assessment. The technique can be combined with an acetazolamide challenge to assess the cerebrovascular reserve [42, 43]. The development of selective ASL, which can visualise the vascular territory of a single vessel, is particularly applicable to these investigations [44].

The physiological basis of ASL may however pose difficulties in patients with suspected cerebrovascular disease. Qualitative assessment of cerebral blood flow maps can be clinically useful but may be insufficient and inaccurate in the presence of globally reduced cerebral blood flow, such as in patients with bilateral carotid disease or reduced cardiac output. ATT is well-defined for healthy vasculature but can vary greatly in disease states, possibly making conventional ASL techniques with a single PLD unsuitable [45]. Slow flow is more difficult to quantify than high flow [46]. Observed variability in ATT for chronic stroke patients, for example, indicates that individual transit delay times ought to be established in cases of compromised cerebral vasculature before collecting CBF data [47]. Recent work demonstrated a multi-PLD technique that measures both CBF and ATT, creating parametric maps of regional ATT to better model pathological tissue [48, 49]. The technique was recently validated against gold standard H2[15O]-PET measurements [43] and DSC [50] in instances of steno-occlusive disease.

Diagnosis and evaluation of neurodegenerative disease is a rapidly expanding application of ASL, particularly with recent growth in biomarker exploration for Alzheimer’s disease (AD) and dementia [108]. Not only does the quantification of perfusion offer a role in monitoring disease progression, which is potentially useful for clinical trials of new therapies, it may also be able to provide helpful diagnostic information if characteristic perfusion patterns can be established. Changes in brain metabolism and perfusion often precede observable structural changes such as atrophy in neurodegenerative diseases. This is most commonly assessed with fluor-deoxyglucose (FDG)-PET, which is not always cost-effective [109] and has limited availability, in contrast to ASL, which can be added to the routine structural MRI examination at low incremental cost and only 5 min of additional scan time. CBF measurements derived from ASL correlate with gold standard H2[15O]-PET [110, 111].

AD is the most common cause of dementia [108]. The use of ASL in AD was recently reviewed [112], assessing the technique as an emerging biomarker. Altered metabolic activity measured with FDG-PET has been linked to early pathologic changes in AD patients in multiple studies, and regions of altered perfusion measured with ASL appear to overlap with these findings [113, 114]. A number of studies have yielded interesting perfusion comparisons between patients with various forms of dementia and demographically matched healthy controls. ASL has detected global perfusion decreases in AD patients [115] in addition to regional hypoperfusion. Both AD and mild cognitive impairment (MCI), the intermediate stage between normal cognitive decline and dementia, have been associated with hypoperfusion in the middle occipital areas, medial temporal lobe, and especially the parietal lobe [116]. Similar hypoperfusion has been reported in the posterior cingulate and precuneus in addition to certain frontal and parietal regions [117, 118]. Frontotemporal dementia is also associated with regional hypoperfusion, though the spatial distribution is different from that of AD [119], with bilateral hypoperfusion in the frontal cortex and insula and hyperperfusion in the precuneus and posterior cingulate. ASL is potentially a useful tool for this important differential diagnosis [120, 121].

Regional hypoperfusion as assessed with ASL may be a good predictor of cognitive decline in AD and potentially identify candidates for treatment trials [122]. Regional reductions in CBF have also been correlated with known AD risk factors like the apolipoprotein E epsilon 4 allele, further supporting ASL’s utility in indicating disease progression from MCI to AD [123]. Findings from Dashjamts et al. suggest that ASL imaging is better than morphological imaging (voxel-based grey matter density) in the diagnosis of AD [124], although Bron et al. report no added diagnostic value of ASL over structural MRI atrophy markers in presenile dementia classification [125]. A number of studies utilise ASL as part of a multi-tool approach for elucidating AD-related changes, for instance alongside magnetic resonance spectroscopy [126] or hippocampal volume measurement [127]. Regional reduction of perfusion in AD is not necessarily related to other instances of neuropathological hypoperfusion. Proposed mechanisms for AD and subcortical ischaemic vascular dementia remain controversial [128], with debate as to whether hypoperfusion is a cause or consequence of neurodegeneration [115]. Yoshiura et al. found no correlation between regional ATT prolongation and regional hypoperfusion in AD [129], which suggests a different mechanism than that of cerebrovascular disease. While co-occurring pathologies make it difficult for researchers to agree, vascular abnormalities have recently been suggested to play a critical role in AD pathology given known characteristics of beta-amyloid clearance dysfunction [130].

As biomarker research has greatly expanded over the last two decades, reconciliation of the accumulating findings comes with many considerations [108], particularly the translation of group-level findings to individual patients [131]. Overall, the literature demonstrates both the technical feasibility and advantages of ASL in neurodegenerative disease. Evidence suggests that AD staging models ought to incorporate CBF changes as an early biomarker [132]. The relative low cost and ease of implementation of ASL favour its potential inclusion in methods for long-term surveillance of ageing populations. Wang et al. demonstrated the feasibility and compelling clinical utility of ASL across multiple sites [133].

Perfusion imaging provides useful information about vascularisation and vascular proliferation, which is directly applicable to the assessment of brain tumours. Tumour growth requires a substantial supply of blood. Angiogenesis occurs once neoplastic tissue reaches a critical mass, and the particular function and architecture of new blood vessels are linked to tumour type [142]. While relative cerebral blood volume (CBV) measured by DSC perfusion MR imaging is the most common measurement, CBF measured by ASL perfusion imaging has comparable utility in tumour diagnosis, grading, and follow-up of tumours and treatment.

While it may not be appropriate to compare perfusion values determined with different techniques [142], multiple studies seem to indicate that DSC and ASL findings correlate well. Although DSC perfusion is more widely used to evaluate brain tumours, Lehmann et al. [143] found that pCASL detected gliomas, metastases, and meningiomas on a 3 T scanner as accurately as DSC, and Järnum et al. [144] reported a similar correlation for a variety of tumour classifications. More recently, Hirai et al. [145] found that ASL measurements nearly matched DSC quantifications of regional CBF in gliomas, and van Westen et al. [146] reported ASL and DSC blood volume correlation for various intracranial tumours.

Pretreatment classification of tumour types can be hugely important for timely and effective intervention. A recent study found that ASL performed as well as DWI and FDG-PET in distinguishing primary central nervous system lymphomas (PCNSLs) from glioblastoma, identifying the denser and less vascular PCNSLs due to decreased perfusion [147]. This example demonstrates the strategy for using ASL to differentiate tumour types: correlating known histological characteristics with perfusion measurements to yield information otherwise collected by more invasive means. A number of other studies have illustrated this approach. Noguchi et al. [148] found statistically significant differences in ASL-quantified tumour perfusion when comparing haemangioblastomas to gliomas, meningiomas, and schwannomas due to the tumour types’ different vascular densities. Three-tesla PASL has been used to differentiate haemangioblastomas and metastatic tumours, which are the major differential diagnoses of tumours in the posterior fossa in adults [149]. Recently, PASL showed high accuracy in distinguishing pilomyxoid astrocytomas (a relatively new WHO Classification addition) from pilocytic astrocytomas [150]. This type of differentiation is essential for making therapeutic decisions.

Tumour grading is another important informant of treatment and is partly determined by the extent of vascular proliferation. High- and low-grade gliomas can be distinguished on the basis of perfusion quantification with CASL [151, 152] and PASL [153‐156]. Relative CBF measured with ASL has been shown to correlate with relative CBV measured with DSC [157]. Generally, high-grade tumours exhibit CBF above the individual mean, while low-grade tumours exhibit CBF below the mean [153].

Follow-up of tumours with perfusion imaging can greatly inform management. Adult low-grade tumours undergo malignant transformation, which is characterised by a switch from avascular to vascular tumour, known as the angiogenic switch. This is yet to be measured with ASL. Perfusion in high-grade tumours can be measured with ASL before and after treatment to assess response and progression. This is useful in the context of anti-angiogenic treatment: in a case report of a patient treated for recurrent glioblastoma, ASL illustrated tumour progression before conventional MRI did [158]. ASL was furthermore reported to be more effective than DSC at distinguishing radiotherapy-induced necrosis from high-grade glioma recurrence (sensitivity >90 %) [159].

As with any imaging exam, quality control is essential prior to clinical assessment. Common artefacts include motion, signal dropout, distortion, bright spots, and labeling failure [160]. Motion artefacts may appear as rings or curved lines and may result in artefactually high or low CBF values (Fig. 12). Signal dropout and distortion result from susceptibility effects with EPI-based readout sequences. These typically occur at air–tissue interfaces, such as near the frontal sinuses or mastoid bone. Metallic surgical material and haemorrhage are additional sources of such artefacts. Bright spots are random—clusters of—voxels of very high perfusion, due to residual vascular signal. As described above, note that this artefact is in fact of diagnostic value as a feature of slow flow in acute ischaemic stroke. Failure to label the inflowing blood, e.g. due to local susceptibility artefacts, results in apparent lack of perfusion in the entire affected vascular territory (Fig. 7). Finally, ASL should never be acquired after the administration of gadolinium-based contrast, as the resulting T1-shortening is detrimental for the label (Fig. 13). Note that this effect is noticeable for days after contrast administration.

Brain perfusion is highly dynamic and will be influenced by many factors [161]. Some factors are likely to generate long-lasting global and local perfusion adaptations. These include genetics, cognitive capacity [162], personality traits [163], physical exercise [164], and age [165]. Other factors are much more variable and define a state of mind at the time of examination. These include for instance mood [166], blood gases [167], nutrition [168], stress [169], and medication use. Some of the induced changes are global and related to vascular tonus, while other local variations are the results of psychotropic effects on the brain. For the interpretation of perfusion maps and derived values, such variance should ideally be taken into account. Some changes induced by physiology, state of mind, and non-metabolism-related factors may confound interpretation related to disease states, because such changes are in the same areas and of a comparable magnitude. Especially early abnormalities might be obscured by the noise introduced by confounds and modifiers of perfusion. It is therefore suggested that studies of perfusion of the brain always inquire about possible confounds and modifiers using a questionnaire. This information can then be used to understand the validity of perfusion alterations. Where possible, corrections can be applied to the quantitative results by measuring the modifier, e.g. blood gases and haematocrit. The use of a standard operating procedure while performing the perfusion measurement is also advisable to reduce the influence of, e.g. time of day, wakefulness/consciousness, satiety, acute substance use, and the use of certain prescription drugs. As an example, drinking a cup of coffee or smoking a cigarette just before the perfusion measurement has substantial influence on both global and local quantification [170, 171].

As previously mentioned, ASL is particularly effective for imaging paediatric populations. Non-invasive imaging avoids ethical controversy over the injection of contrast agents, and children have a higher baseline CBF signal, which results in a better signal-to-noise ratio [4]. Repeat measurements, the accuracy of which has been experimentally demonstrated, are useful in neurodevelopmental studies of children [172] and sustained monitoring in high-risk neonates, particularly in attempts to identify factors linked to brain injury in development [173]. Though applications of ASL in epilepsy, neoplasms, and other neurological disorders have not been fully evaluated for paediatric groups, the same principles apply. Studies have demonstrated the utility of ASL in evaluating perfusion changes associated with sickle cell disease, the most common cause of stroke in children [174]. However, care must be taken to not overestimate CBF abnormalities. Since most ASL studies are done in adult populations, sequences must sometimes be optimised for a younger population with different average physiological parameters [175].

ASL’s ability to quantify CBF is its major advantage over contrast-enhanced MR perfusion techniques. Quantitative assessment of CBF potentially allows for the use of threshold or reference values, thus aiding in the—more objective—differentiation between healthy and diseased conditions. Such an assessment is less user- and experience-dependent and more sensitive to subtle alterations than visual assessment.

CBF is commonly quantified according to the following equation for (p)CASL [21]:where λ is the brain/blood partition coefficient in millilitres per gram, α is the labeling efficiency, M _c is the control and M _l is the labeled signal intensity, and M _PD is the signal intensity of a proton density-weighted image. This is used to normalise the overall signal intensity which potentially varies due to several confounders, such as hardware and patient variability globally affecting the signal. Recommendations on how to obtain M _PD can be found in Alsop et al. [16]. Six thousand is a conversion factor of the units from ml/g/s to ml/100 g/min. While many of the parameters can be estimated, it is common to use nominal values indicating that quantification is still partially based on assumptions. Nominal values are as follows: λ = 0.9 ml/g [176], T _1Blood = 1,650 ms at 3.0 T [177] and 1,350 ms at 1.5 T [178], α = 0.85 for pCASL [179], and 0.98 for PASL [180].

An important aspect affecting the quantitative assessment of CBF is the partial volume effect, related to the fact that the voxel size of ASL is several times that of 3D T1-weighted acquisitions (Fig. 14). Hence, most voxels contain a combination of tissues and/or cerebrospinal fluid. Different tissues have substantially different perfusion characteristics. Perfusion in the white matter is about half that of grey matter, while at the same time T1 is shorter and arterial arrival time is longer. The partial volume effect becomes even more relevant in the presence of atrophy, when voxels contain relatively less grey matter and subsequently lower average CBF. Such unwanted effects are dealt with by using a post-processing technique called partial volume correction [181].

In the evaluation of epilepsy, SPECT and PET are used to measure ictal and interictal blood flow for localisation of epileptogenic activity, illustrating increased perfusion in the critical period and decreased perfusion after seizures [194]. However, nuclear medicine techniques can be expensive or contraindicated. ASL can potentially assist clinical diagnosis by quickly identifying potential perfusion asymmetries or ruling out causes such as stroke. Non-invasive measurements could be repeated to assess changes over time, particularly in response to treatment. Repeated seizures can lead to a chronic epileptic condition, demonstrated experimentally [195] and clinically [196], and it is thought that neuroinflammation and seizure-induced vascular alterations contribute to epileptogenesis [197], warranting investigation of such changes with perfusion imaging.

A few studies have employed interictal [198‐200] and immediate post-ictal [201, 202] PASL measurements in cases of partial epilepsy. Correlation between ASL, PET, and electrophysiological data has been shown in patients with interictal hypoperfusion in surgically intractable epilepsy [203] and tuberous sclerosis [204]. PASL also compares favourably to PET and electrical source imaging in identifying the epileptogenic zone [205], indicating diagnostic value and potential utility in guiding intracranial electrode placement. Recent work demonstrated comparable measurements between PASL and DSC for detecting seizure-associated perfusion asymmetries [206].

Ictal ASL measurements have only been reported in case studies, as seizures must coincidentally occur during scans. One patient’s complex partial seizure during PASL acquisition showed ictal hyperperfusion in the typical seizure focus, a finding that was not present interictally [207]. This matches the underlying pathophysiological mechanism of a seizure; excess activation requires increased blood flow to deliver glucose and oxygen. Oishi et al. [208] captured PASL images of a patient during and after partial status epilepticus. The case study followed the same trends, with marked hyperperfusion in the epileptogenic zone during the seizures and hypoperfusion compared to the non-affected hemisphere once the seizures stopped. Finally, Kanzawa et al. report that the combined use of ASL and DWI was useful in differentiating non-convulsive partial status epilepticus from stroke when ictal discharges were not captured with EEG, leading to appropriate treatment and resolution of symptoms [209].

Ictal perfusion imaging provides better information than interictal imaging alone [210]. Since ictal SPECT only depends on injection time whereas ictal ASL requires simultaneous scanning, which is not typically compatible with electroencephalography, regular utility of ASL is currently limited to interictal foci lateralisation. Perfusion quantification can also yield valuable insight into seizure-causing lesions such as cortical dysplasia [211]. Identification of such lesions is vital for pre-surgical planning and operative outcome.

The use of ASL in exploring the neurobiological and anatomical correlates of clinical psychiatric conditions is steadily increasing. It is thought that cerebral microvasculature abnormalities may play a role in some psychiatric illnesses, the mechanisms of which are poorly understood [212], but changes in regional CBF are generally attributed to changes in neuronal activity and energy demand secondary to psychiatric disease or its therapy.

Baseline differences in CBF between patients and healthy controls have been identified in several ASL studies. Chronic and treatment-resistant depression showed associated perfusion abnormalities in the pathologically relevant subgenual anterior cingulate cortex compared to healthy controls [213]. Adolescent depression appears to affect regional CBF measurements in executive, affective, and motor networks [214]. Late-life depression has been associated with elevation of white matter CBF [215]. Schizophrenia has been correlated with hypoperfusion of the prefrontal cortex [216‐218], in addition to complex patterns of altered perfusion in other brain areas. Borderline personality disorder was associated with decreased CBF in the medial orbitofrontal cortex and increased CBF in the lateral orbitofrontal cortex compared to healthy controls [219].

ASL may also be used to assess treatments. Regional CBF measurements have been shown to predict response of patients with depression [220] and schizophrenia [221] to repetitive transcranial magnetic stimulation, indicating potential for individualised therapy. Studies isolating the effects of pharmaceuticals on brain perfusion in healthy subjects allow for greater understanding of their application in psychiatric conditions, where changes in regional perfusion can complicate such assessment [222‐224].

Territorial ASL refers to the ASL method that provides a perfusion measurement from a specific vascular territory (Fig. 15). The original methods, based on PASL sequences, allowed for a separation mostly of each of the two perfusion territories from the internal carotid arteries (ICAs), and the posterior territory, without distinction between ACA and MCA territories or between left and right vertebral arteries [225]. More recent methods, based on pCASL, allow for a much finer assessment of the perfusion territories from smaller arteries, with the caveat that separate measurements from the ACA and MCA will require imaging of perfusion in tissues distal to the circle of Willis and will therefore never be able to produce whole brain perfusion maps [226]. For an extensive review on all territorial ASL methods, the reader is directed to Hartkamp et al. [227]. Territorial ASL has been used in many studies of sub-acute stroke and has shown to provide additional information on the underlying pathology of lateralised lesions. In particular, it has been shown that in about 10 % of patients with a cortical lesion, assessment of the anatomical location alone results in a misclassification of the perfusion territory in which the infarct is located. The availability of a vascular territorial perfusion map allows to correct this [82]. An additional potential application of territorial ASL is in the context of steno-occlusive disease, when both information on the affected vessel’s perfusion territory and an assessment of cerebrovascular reserve capacity can be obtained.