How far is arterial spin labeling MRI from a clinical reality? Insights from arterial spin labeling comparative studies in Alzheimer's disease and other neurological disorders.

Abstract

Clinical imaging such as FDG-PET and SPECT has been well established in the diagnosis of Alzheimer's disease (AD) and other neurological disorders, but the diagnostic value of arterial spin labeling (ASL) MRI, a relatively new noninvasive imaging technique, has not been validated or confirmed. Comparative studies that compared ASL with FDG-PET, 15O-water-PET, SPECT, dynamic susceptibility contrast-enhanced (DSC), or dynamic contrast-enhanced (DCE) MRI are important to reveal the concordance and discordance between ASL and these clinical imaging tests. This paper reviews such studies in healthy subjects and patients with AD or other neurological disorders. These comparative studies have demonstrated that ASL and FDG-PET/15O-water-PET had overall concordance and regional variability in healthy subjects, and the regional hypoperfusion that ASL depicted in diseases such as AD was largely concordant with the abnormalities identified by FDG-PET/SPECT (the diagnostic performance was comparable), which suggests that ASL may have a similar diagnostic value as FDG-PET/SPECT. In addition, there are mixed findings in ASL cerebral blood flow (CBF) measurements compared with 15O-water-PET, and ASL may be less sensitive than DSC (or DCE) in the diagnosis of neurological disorders such as stroke. Studies with large samples are needed to further validate the diagnostic value of ASL and its CBF measurement. Furthermore, it is necessary to optimize and standardize ASL parameters and explore new techniques to overcome the limitations of ASL and make it a clinical imaging alternative to invasive FDG-PET/SPECT/DSC/DCE in the diagnosis of AD and other neurological disorders. IT IS OFTEN difficult to differentiate early-stage Alzheimer's disease (AD) from other neurodegenerative dementias or mild cognitive impairment (MCI).1 MRI, FDG-PET, and SPECT have been the major clinical imaging tests used for the diagnosis of dementia. FDG-PET or SPECT uses radioactive tracers to image glucose metabolism or brain perfusion. Since reductions in cerebral glucose uptake and brain perfusion often precede brain atrophy, measures of FDG-PET/SPECT can serve as early biomarkers for preclinical or mild AD.2 FDG-PET/SPECT imaging in neurodegenerative disorders is based on the coupling of neural activity and glucose metabolism to reduced regional cerebral blood flow (CBF)3 due to the synchronization of cerebral metabolic demands to regional blood supply. Thus, spatial patterns of regional CBF in perfusion imaging tend to mirror FDG-PET glucose metabolism results unless neurovascular coupling is compromised.4 In addition, CBF measurements are important in the assessment of AD and other neurological disorders such as stroke. 15O-water-PET (PET imaging with 15O-water tracer) is considered a gold standard for quantifying CBF.5 The diagnostic value of PET and SPECT imaging has been well demonstrated and validated,1, 3 but they are invasive and expensive, and need exposure to radiation. Although relatively new, arterial spin labeling (ASL) MRI has emerged as a perfusion imaging technique of growing popularity due to its recent technical advancements and use of magnetically labeled arterial blood water as an intrinsic tracer. It measures CBF by computing the difference between the labeled and nonlabeled images. ASL is noninvasive, less expensive, convenient for acquisition (can be acquired in the same section of structural MRI) and allows multiple repeated measurements. Thus, it might be an alternative to invasive imaging tests such as FDG-PET and SPECT. There are mainly three types of ASL: continuous ASL (CASL), pulsed ASL (PASL), and pseudo-continuous ASL (pCASL). They are different in the way they label arterial blood water: CASL uses a single long pulse, PASL uses one or several short pulses, while pCASL applies over 1000 shaped radiofrequency pulses at around one per millisecond. Among the three, pCASL has the highest signal-to-noise ratio and superior labeling efficiency. Therefore, pCASL is recommended for clinical imaging.6 ASL studies have found that areas of hypoperfusion in dementia such as AD7-13 and frontotemporal dementia (FTD)10-15 are similar to those identified by FDG-PET or SPECT.7, 11, 16-21 It is of clinical interest to assess the diagnostic value of ASL due to its ease of access, noninvasiveness, cost-effectiveness, and free of radiation exposure. In addition, dynamic susceptibility contrast-enhanced (DSC) MRI is widely used for detecting perfusion deficits in neurological disorders such as stroke and brain tumors. It has high signal-to-noise ratio (SNR) and short scan time (1 min, due to T2*-weighted fast EPI sequences), but it is invasive and requires injection of contrast materials. The contrast material gadolinium could induce nephrogenic systemic fibrosis in patients with renovascular disease22 and DSC is sensitive to susceptibility artifacts (such as hemorrhages and calcifications), which limits its applicability. Another perfusion MR imaging technique, T1-weighted dynamic contrast-enhanced (DCE) MRI is more robust to susceptibility artifacts and has higher spatial resolution, but also needs injection of contrast materials. In contrast, ASL is noninvasive and insensitive to susceptibility artifacts (such as hemorrhages), and thus it might be an alternative to DSC/DCE MRI for detecting perfusion deficits in neurological disorders such as stroke and brain tumors. Over a decade, ASL has been used to study AD and several other neurological conditions such as cerebrovascular disorders and brain tumors.23 For a comparative overview of PET, SPECT, DSC, ASL and other brain perfusion imaging techniques (including detailed description and imaging parameters or features such as spatial resolution, reproducibility, accuracy, and scan time for each imaging technique), see the comprehensive review by Wintermark et al.23 While FDG-PET, SPECT, and DSC/DCE are well established clinical imaging tests, ASL is not, mainly due to its unconfirmed or invalidated diagnostic value. In recent years, an initiative has started to address the final steps to make ASL a clinical reality,24 and recommendations of ASL implementation for clinical applications have been made.6 However, how far is ASL from a clinical reality? Studies that compared ASL with 15O-water-PET, FDG-PET, SPECT, DSC, or DCE MRI are important to reveal the concordance and discordance between ASL and these clinical imaging tests, and the diagnostic value of ASL. This paper reviews comparative studies between ASL and 15O-water-PET/FDG-PET/SPECT/DSC/DCE in healthy subjects and patients with AD or other neurological disorders to assess the diagnostic value of ASL and understand how far ASL is from a clinical reality. This section is organized as follows: (1) Validation of ASL against other imaging methods in the healthy brain, and (2) ASL comparative studies in the disease states. The disease states in the latter mainly include neurodegenerative disorders (e.g., AD), cerebrovascular disorders, and brain tumors. ASL has been compared with 15O-water-PET, FDG-PET, and DSC MRI in the healthy brain, and several studies have attempted to validate ASL against the gold-standard 15O-water-PET for CBF measurements.25-32 Findings of validation against 15O-water-PET indicate either insignificant difference in (average whole brain or gray matter) CBF measurements using the two imaging methods,25, 27, 30 or significantly higher ASL-CBF values than those of PET.29, 31 Table 1 presents the details of these studies. Briefly, using 15O-water-PET as the gold standard for CBF quantification, Ye et al conducted the first ASL comparative study in 12 healthy subjects and found insignificant difference between ASL- or PET-measured CBF values in the cortical strips, but significantly lower ASL-measured CBF value in a central white matter region which may be due to longer arterial transit time (ATT) in regions of white matter.25 Later, Qiu et al assessed ATT of ASL in healthy subjects (n = 14) and showed changes of ATT could lead to regional CBF difference between ASL and 15O-water PET.26 Henriksen et al further evaluated PASL, 15O-water PET, DCE MRI and phase contrast mapping (PCM) imaging in healthy subjects (n = 17), and reported insignificant difference in the average global CBF values measured by ASL, PET, and DCE (e.g., ASL-CBF slightly lower than PET-CBF) (Fig. 1).27 In addition, Arbelaez et al examined healthy subjects (n = 9) during hypoglycemia and euglycemia with PASL and 15O-water PET, and reported similar increases in regional CBF on ASL and PET in response to hypoglycemia.28 Van Golen et al further found correlated global CBF values between pCASL and 15O-water PET in healthy subjects (n = 11) and type 1 diabetic patients (n = 20), but significantly higher ASL-CBF values than PET-CBF.29 In a recent study, Heijtel et al demonstrated that pCASL CBF imaging had comparable precision (in CBF measurements) as 15O-water PET in healthy subjects (n = 16) during baseline and hypercapnia, although there were areas with a CBF discrepancy such as the overestimated (e.g., deep cortical tissues) and underestimated (e.g., the basal nuclei) regions on ASL (Fig. 2).30 Furthermore, Zhang et al compared simultaneously acquired 15O-water PET and pCASL in healthy subjects (n = 10) and found overall similarity (correlation coefficient r = 0.81–0.88) between ASL and PET with significant correlations in 9 of 10 regions of interest (ROIs).31 They also found significantly higher ASL-CBF values than PET-CBF, and regional and individual differences between the two imaging modalities (e.g., higher CBF in the mid-frontal and posterior cingulate regions on ASL).31 In another study using simultaneously acquired PASL and 15O-water PET, Andersen et al found that global ASL CBF and PET CBF were concordant at baseline but not during hyperperfusion (at acetazolamide stimulus) in newborn piglets (n = 7).32 Cerebral blood flow maps by various methods. Co-registered CBF maps from three different subjects (before pixels containing vessels are removed) and the corresponding T1-weighted 3D anatomical scan. All CBF maps are identically scaled (0–100 mL/100 g/min). Estimated global CBF is given below each CBF map. CBF = cerebral blood flow; DCE = dynamic contrast enhanced T1-weighted perfusion magnetic resonance imaging; ASL = arterial spin labeling; PET = positron emission tomography. (Henriksen et al., 2012; courtesy of Dr. Henriksen; reprinted with permission from the publisher of Henriksen et al., 2012.) The average CBF maps for 15O H2O PET (A,G), pCASL (B,G), and pCASLcrush (C) during baseline (A–C) and hypercapnia (F–G) combined with voxel-wise parametric testing results (P < 0.001, uncorrected) between 15O H2O PET–pCASL (D,H) and 15O H2O PET–pCASLcrush (E). (Heijtel et al., 2014; courtesy of Dr. Heijtel; reprinted with permission from the publisher of Heijtel et al., 2014.) Taken together, despite the overall similarity between ASL and 15O-water PET imaging, the CBF values measured by ASL vary from slightly lower than PET-CBF25, 27 to significantly higher than PET-CBF.29, 31 Higher ASL-CBF values may be caused by the overestimation of intravascular flow signal31 which can be corrected by using vascular crushing gradients,30 while lower ASL-measured CBF values may be due to longer ATT in certain brain regions which can be corrected by adjusting ASL parameters such as the inversion time or postlabeling delay. Overall similarity and concordance has also been reported in ASL studies against FDG-PET in the healthy brain.33-35 Newberg et al measured concurrent changes of CBF and glucose metabolism during a visual stimulation task in healthy subjects (n = 5) and found an excellent concordance between regional CBF and regional cerebral metabolism rate.33 Cha et al further compared cerebral perfusion and glucose metabolism at resting-state in healthy subjects (n = 20) using pCASL and FDG-PET, and reported good overall correlation between ASL and FDG-PET, as well as regional variability and discordance (e.g., in the medial temporal regions such as the hippocampus and amygdala).34 The overall significant correlation between ASL and FDG-PET measurements with regional variations (lowest correlation in regions such as the hippocampus and amygdala) has been confirmed in oncology patients who had no neurological disorders (i.e., the subjects had normal brains) (n = 13) by Anazodo et al using simultaneously acquired ASL and FDG-PET images.35 DSC MRI is considered the clinical standard for perfusion MR imaging in neurological disorders such as stroke and brain tumors,23 and CBF values measured by DSC are comparable to those by 15O-water PET or xenon CT.36, 37 Li et al and Weber et al found that CBF values obtained by ASL and DSC were comparable in normal brain tissue.38, 39 Details of these studies are listed in Table 1. Taken together, the validation studies that compared ASL with 15O-water-PET, FDG-PET, or DSC MRI indicate that ASL and these clinical imaging tests had overall similarities and regional differences in the healthy brain. However, there are mixed findings in the CBF measurements using ASL and 15O-water-PET.25, 27, 29-31 Studies with larger samples are needed to further validate ASL CBF quantification and improve it. Comparative studies between ASL and 15O-water PET/FDG-PET/SPECT have investigated functional abnormalities in neurodegenerative disorders such as AD and FTD.7, 11, 40-44 Correlated CBF measurements using ASL and 15O-water-PET have been reported in AD and MCI.40, 41 Xu et al examined 2 patients with MCI, 1 patient with AD, and 19 healthy subjects with ASL and 15O-water-PET, and found good global and regional correlation in the gray matter and posterior cingulate cortex (r = 0.74-0.79).40 Kilroy et al conducted a longitudinal study for patients with MCI (n = 6) and AD (n = 1) using 2D EPI ASL, 3D GRASE ASL, and 15O-water PET, and reported that GRASE ASL was more spatially correlated with PET (r = 0.45), and more longitudinally repeatable than EPI ASL.41 Table 2 shows the details of these studies. Similar patterns of regional abnormalities and comparable diagnostic performance between ASL and FDG-PET/SPECT have been found in patients with AD or FTD.7, 11, 42-44 Using a composite region of interest (ROI) approach, Chen et al compared pCASL and FDG-PET in patients with probable AD (n = 15) and reported concordance between the patterns of hypoperfusion and hypometabolism in patients and comparable diagnostic accuracy: areas under the receiver operating characteristic curves (AUCs of ROCs) were 0.94 for ASL and 0.92 for FDG-PET.42 In addition, Musiek et al compared concurrently acquired pCASL and FDG-PET images in patients with AD (n = 15) and demonstrated that ASL and FDG-PET identified similar regional deficits and had comparable diagnostic performance (AUCs of ROC: 0.91 for ASL and 0.90 for FDG-PET), although some artifacts were observed on ASL.43 Furthermore, Verfaillie et al examined 18 AD patients and 12 FTD patients with pCASL and FDG-PET, and found similar patterns of hypometabolism and hypoperfusion, and comparable diagnostic performance of metabolism and perfusion in regions typical to AD and FTD (Fig. 3).11 In early-onset AD, Verclytte et al found that ASL and FDG-PET identified similar pathological regions such as the inferior parietal lobules and the temporal cortex in patients (n = 37).44 In addition, Takahashi et al examined patients with suspected AD (n = 68, 32 of them with non-AD) with pCASL and SPECT, and demonstrated high concordance between the two imaging modalities inspected by two radiologists.7 They reported higher AUC of ROC of SPECT than that of ASL (SPECT versus ASL: ∼0.8 versus 0.69) mainly due to the watershed artifact on ASL, and concluded that compared with SPECT, the diagnostic performance of ASL was slightly (but insignificantly) lower in the differential diagnosis of AD and non-AD.7 Transversal FDG and ASL images of an FTD (first and second rows, MMSE 26) and an AD (third and fourth rows, MMSE 17) patient with early-onset disease. Both transversal planes show predominantly prefrontal abnormalities in FTD and parietal abnormalities in AD. Red color reflects normal metabolism and perfusion. (Verfaillie et al., 2014; courtesy of Dr. Verfaillie; reprinted with permission from the publisher of Verfaillie et al., 2014.) Despite the strong concordance and correlation between hypoperfusion and hypometabolism measurements by the two imaging techniques in AD and other neurodegenerative disorders, there are some discordance. For example, hyperperfusion in the medial temporal regions has been found on ASL,45 but hypometabolism in these regions on FDG-PET in AD patients.46 Another example, ASL revealed regional hyperperfusion in areas such as medial parietal cortex, and posterior cingulate in FTD,14 which has not been found on FDG-PET. Taken together, correlation between ASL and 15O-water-PET measures has been found in patients with AD or MCI.40, 41 Studies also found that ASL and FDG-PET provided largely overlapping information and comparable diagnostic performance in patients with neurodegenerative dementias such as AD and FTD. However, ASL is more influenced by hemodynamic factors such as ATT than FDG-PET or SPECT which leads to artifacts on ASL7, 43 and reduces its diagnostic performance. Minimizing these artifacts will improve the diagnostic accuracy of ASL and make it a promising alternative to clinical PET or SPECT imaging. ASL has been compared with the gold standard 15O-water-PET imaging in cerebrovascular diseases and it has been found that CBF values measured by ASL and 15O-water-PET are correlated, but ASL-CBF values are higher.47, 48 Due to the usually long ATT in patients with such disorders, adjustment of ASL parameters such as inversion time (TI) or postlabeling delay (PLD) is necessary to reduce arterial transit delay artifacts on ASL images. Using multiple TIs, Bokkers et al observed significantly decreased CBF in the middle cerebral artery territory ipsilateral to the lesion site on both ASL and 15O-water-PET images, correlated gray matter ASL-CBF and PET-CBF values (r = 0.58), and higher average gray matter (aGM) ASL-CBF than that of PET in patients with symptomatic carotid artery occlusion (n = 14).47 In addition, correlation between CBF values measured by PASL and 15O-water-PET (r = 0.52–0.69), slightly higher average ASL-rCBF (relative CBF) than PET-rCBF, and correlation between ASL-ATT and PET-MTT (mean transit time) were found in patients with steno-occlusive diseases (n = 16).48 The correlation between ASL and DSC measurements in cerebrovascular diseases has varied from modest49-51 to moderate-or-good.52-58 Siewert et al found agreement between the two imaging techniques in 17 of 21 (81%) patients with stroke, and that PASL could detect perfusion abnormalities in stroke similar to DSC MRI.59 Wolf et al also found that after counting out the ATT difference between groups, the CBF measurements using CASL and DSC MRI were correlated in patients with cerebrovascular disease (n = 11), and concluded that CASL and DSC MRI might be complementary for examining perfusion in cerebrovascular disease.60 Later, Wolf et al applied PASL with multiple inflow times to ASL-and-DSC-comparison in patients with ischemic stroke and reported significant correlation (r = 0.42–0.49) between ASL perfusion maps and DSC perfusion imaging (Fig. 4).61 Using rat stroke models (n = 12), Tanaka et al found that CBF values obtained from ASL and DSC were comparable in normal brain tissue, but in tissue with postischemic hyperperfusion, CBF values from the two imaging techniques differed significantly due to ischemia induced changes.62 Wang et al further pointed out that pCASL and DSC MRI could depict regions of hypoperfusion in patients with acute stroke (n = 26) consistently with significant correlation (r = 0.79) between ASL and DSC-mean-transit-time (MTT), and pCASL was better at depicting regions of hyperfusion than DSC MRI.55 For perfusion deficit detection, Bokkers et al found that in patients with acute stroke whose perfusion deficits were detected by DSC (n = 39), pCASL detected 32 (missing 7 lesions, concordance 82%); and compared with DSC (that could detected smaller perfusion deficits), pCASL could depict large perfusion deficits and perfusion-diffusion mismatches (region of perfusion abnormality larger than diffusion abnormality).63 Zaharchuk et al compared the ASL-diffusion-weighted imaging (DWI) and DSC-DWI mismatch classification in patients with stroke (n = 43) and found that ASL and DSC frequently agreed (57–77%), but ASL tended to overestimate perfusion lesion size.49 In addition, Nael et al reported that the concordance of pCASL and DSC in detecting perfusion abnormalities were found in 71–80% of the cases (ASL failed to detect perfusion deficit in 11% of patients), and pCASL was less sensitive than DSC in detecting regional CBF changes (especially hyperperfusion after recanalization) in acute ischemic syndrome (n = 25).52 Furthermore, Nael et al found that pCASL overestimated the hypoperfusion volume in acute ischemic syndrome (n = 41) and the diffusion-perfusion mismatch tissue classification, and factors such as arterial transit delay, flow pattern, and velocity changes could affect the performance of ASL.53 However, recent studies show that ASL with multiple inflow times can improve the identification of regions of delayed ATT and reduced CBF.54, 61 A 72-year-old patient with acute ischemic stroke of the left MCA territory (NIHSS 6) and corresponding distal stenosis of the left MCA (see DWI and TOF-MRA, box A). DSC PWI shows a delay of contrast arrival on TTP (A), an increased mean transit time (MTT), and a decrease of the regional cerebral blood flow (CBF) (B). ASL imaging (C) demonstrates a strongly diminished signal intensity reflecting hypoperfusion. (Wolf et al., 2014; courtesy of Drs. Kern and Wolf; reprinted with permission from the publisher of Wolf et al., 2014.) Studies of ASL and DCE MRI comparison in cerebrovascular diseases are few, and the findings of ASL and DCE CBF measurements are mixed: correlation results vary from no correlation (r = 0.24)64 to good correlation (r = 0.74).65 More studies are needed to understand the concordance and discordance between the two imaging techniques. In addition, ASL has been compared with SPECT in cerebrovascular diseases, and correlated ASL and SPECT CBF measurements have been reported.66, 67, 69-71 Arbab et al reported significant positive correlation between regional ASL-CBF and SPECT-CBF in patients with carotid stenosis (n = 9) and using SPECT as reference (100%), the accuracy of ASL-CBF was 80–82%.66 Later, Uchihashi et al found significantly correlated regional CBF and cerebral vasoreativity values measured by PASL and SPECT in patients with carotid stenosis (Fig. 5) (n = 20), and regional perfusion deficits detected by ASL (98–99% accurate) equivalent to SPECT, although ASL might overestimate CBF in hyperperfusion regions.67 Furthermore, concordance and discordance of ASL and SPECT were observed in patients with intracranial dural arteriovenous fistulas (n = 3) (68). In moyamoya disease, Noguchi et al reported significant positive correlation (r = 0.5–0.86) between ASL-CBF and SPECT-CBF (n = 12), suggesting that ASL could detect regions of ischemia as SPECT.69 In a recent study, Noguchi et al found that ASL was equivalent to SPECT (r = 0.56–0.83) in mapping cerebrovascular reserve in patients with this disease (n = 16).70 Furthermore, ASL has been used for postsurgical evaluation to detect hyperperfusion after revascularization and correlation between ASL-CBF and SPECT-CBF derived measurements (asymmetry ratios) (r = 0.8) was found in patients with moyamoya disease (n = 15).71 Typical CBF maps obtained by SPECT and ASL in patients with severe right carotid stenosis. Preoperative resting images (top row), preoperative images with ACZ challenge (middle row), and postoperative images (bottom row) are shown. Preoperative hypoperfusion, poor vasoreactivity, and postoperative hyperperfusion in the right internal carotid artery region can be seen in both SPECT and ASL. (Uchihashi et al., 2014; courtesy of Drs. Hosoda and Uchihashi; reprinted with permission from the publisher of Uchihashi et al., 2014.) Taken together, correlation between CBF measurements using ASL and 15O-water-PET, DSC, DCE or SPECT has been found in patients with cerebrovascular diseases. These comparative studies have demonstrated that ASL may have a similar diagnostic value as SPECT, but it may be less sensitive than DSC or DCE MRI in detecting regional CBF changes in cerebrovascular disorders such as stroke. Techniques such as multiple inflow times can improve ASL imaging and reduce regional CBF underestimation or overestimation of lesion size. The findings of correlation analysis between the measurements using ASL and other imaging techniques (such as 15O-water PET, DSC, and DCE) in brain tumors are mixed, and the correlation results vary from modest72, 73 to moderate-or-good.74-83 Lüdemann et al found that perfusion values of ASL, 15O-water PET, DSC, and DCE in the normal brain tissue and tumor were correlated (r = 0.22–0.49) and ASL-CBF values were higher than PET-CBF in patients with brain tumors (n = 12).72 Roy et al reported that ASL-CBF and DCE-CBF measurements were weakly correlated in gliomas (r = 0.27).73 Warmuth et al found close linear correlation between ASL and DSC tumor-CBF/total-CBF ratios (r = 0.83) (n = 36) and demonstrated that, like DSC, ASL could distinguish between high- and low-grade gliomas.74 Kimura et al reported that CASL and DSC measured tumor-CBF values were positively correlated (r = 0.73), indicating the two perfusion methods were comparable (n = 21).75 Using either PASL or pCASL, Lehmann et al76 (n = 27), Hirai et al77 (n = 24), Jarnum et al78 (n = 28), and Knutsson et al79 (n = 15) confirmed that ASL and DSC had well correlated tumor CBF measurements (r = 0.60–0.89), which suggested that ASL was an alternative to DSC MRI in CBF evaluation of brain tumor. Furthermore, White et al found that although average regional ASL and DSC CBF measurements (in tumor regions) were positively correlated (r = 0.71–0.87), voxel-wise correlation was only in 30–40% of patients (n = 30) indicating the two imaging techniques were regionally similar, but spatially different. This may be due to factors such as the underestimated CBF of ASL in areas with arterial transit delay while unaffected DSC and the assumption in the calculations of DSC CBF.80 The diagnostic performance of ASL in brain tumors has been assessed. Ozsunar et al compared PASL, FDG-PET, and DSC-CBV (cerebral blood volume) in patients with glioma (n = 30) and found that ASL could distinguish recurrent glial tumor from radiation necrosis with relatively high sensitivity and specificity (88%, 89%) [versus FDG-PET (81%, 90%) versus DSC-CBV MRI (86%, 70%)].84 Yamashita et al further evaluated the diagnostic performance of ASL, DWI, and FDG-PET in the diagnosis of brain tumor and found differences in tumor blood flow, diffusion coefficient and FDG uptake measurements that could differentiate primary central nervous system lymphomas (n = 19) from globlastoma multiforms (n = 37) with comparable diagnostic accuracy (ASL CBF: 0.81–0.89; DWI: 0.77; FDG-PET: 0.85).85 In addition, Weber et al examined patients with brain metastases (n = 25) after radiosurgery and demonstrated that the follow-up CBF measurements by PASL and DSC could predict treatment outcome.86 They also found that the colocalization rate of brain tumor spot identified by PASL and DSC was around 83% (n = 22).87 The added value of ASL to DSC has been evaluated in patients with glioblastoma multiforme (n = 62) and Choi et al reported that ASL had similar diagnostic accuracy as DSC (ASL versus DSC: 72.6% versus 75.8%) and it improved the diagnostic accuracy of DSC by 12.9% (DSC and ASL: 88.7%) in differentiating tumor recurrence from pseudoprogression.88 Interestly, Seeger et al also found similar diagnostic accuracy of ASL, DSC, and DCE in distinguishing recurrent gliomas from stable diseases (n = 40): ASL (69.2%) versus DSC (67.5%) versus DCE (65%), and the diagnostic performance improved (82.5%) when combining the three perfusion imaging techniques.89 Taken together, correlation between CBF measurements using ASL and clinical imaging tests (such as 15O-water PET and DSC), and comparable diagnostic performance between ASL and DSC have been found in patients with brain tumors. In addition, the added value of ASL to DSC or DCE in the diagnosis of gliomas has been confirmed. There are few ASL comparative studies in other neurological disorders such as epilepsy. It has been shown that ASL has a value in detecting perfusion deficits for localization or lateralization of seizure focus.90-94 In patients with drug-resistant temporal lobe epilepsy (TLE) (n = 10), it was found that the mean CBF values in the mesial temporal lobe of ASL and 15O-water PET were not significantly different,90 ASL and FDG-PET data were correlated and they were also correlated with the electrophysiological data (n = 3).91 In another study, Storti et al found that seizure source (in the interictal phase) was associated with colocalized regional hypoperfusion and hypometabolism in four of six (67%) patients.92 Blauwblomme et al further examined interictal cerebral perfusion deficits in children with focal cortical dysplasia by means of pCASL and FDG-PET and reported concordance (i.e., colocalized abnormalities) in 5 of 6 (83%) cases.93 Recently, Oner et al evaluated ASL and DSC in detecting perfusion asymmetry for lateralization of seizure focus in refractory TLE (n = 36) and found similar asymmetry indices and similar correlations with clinical lateralization using the two imaging techniques.94 Details of these studies are presented in Table 2. In addition, ASL comparative studies have examined patients with other neurological disorders or neurocognitive impairments,95-97 although such studies are few. Wong compared ASL and DSC (spin-echo SE-DSC and gradient-echo GE-DSC) in patients with various neurological disorders such as brain tumors, abscesses, and hemorrhage, and reported significant correlation of CBF ratios between ASL and the two DSC methods.95 Towgood et al used pCASL and FDG-PET to measure regional glucose metabolism (rCMRglc) and regional CBF (rCBF) in HIV-infected patients (with neurocognitive impairment) (n = 35), and found significant age effects with reduced rCMRglc and rCBF in frontal regions on both PET and ASL, suggesting cross-validity between the two imaging techniques.96 Furthermore, simultaneous PET/MR imaging has been applied to patients with intracranial masses, brain or neck tumors, or neurodegenerative disorders (n = 50), which minimizes temporal variations and allows direct comparison between ASL and PET in these conditions.97 Taken together, the above comparative studies have demonstrated moderate-to-good concordance between ASL and clinical imaging tests (e.g., FDG-PET) in neurological disorders such as epilepsy. Larger studies are needed to further validate the diagnostic value of ASL in these conditions. As a possible alternative to invasive clinical imaging such as FDG-PET and SPECT, ASL has the advantage to identify regional perfusion deficits and quantify regional CBF noninvasively. The comparative studies covered in this review shed insights into the diagnostic value of ASL in AD and other neurological disorders, and the concordance and discordance between ASL and clinical imaging tests. An important finding of the comparative studies is the limitations of ASL in the diagnosis of AD and other neurological disorders (in comparison to clinical imaging). Several factors such as ATT and labeling efficiency influence the quality of ASL images, and their impacts on ASL images vary among different subjects (patients or healthy controls) due to different physiological or pathological conditions. For example, there are delays in the arrival time of arterial blood to the brain in the elderly who have low CBF, which results in arterial transit errors or artifacts on ASL. Such artifacts reduce the diagnostic accuracy of ASL and hinder its clinical usage. Major limitations of ASL are as follows. First, ASL is sensitive to ATT variations. In the healthy brain, ATT may vary between 0.5 and 1.5 s, while in the diseased brain, ATT can be 2 s or longer, and thus, a postlabeling delay (PLD) of 2 s is recommended for clinical adult population.6 This ensures that the PLD is longer than ATT for CBF calculation, but it may be at risk of losing the ASL signal or lowering the ASL signal-to-noise ratio. Longer ATTs in the watershed area may lead to watershed artifact (borderzone sign) in old subjects who have low CBF.7, 43, 98 Takahashi et al found the artifacts in the watershed area of ASL in 70% of non-AD patients (who were misdiagnosed as AD).7 Prolonged ATTs have also been found in patients with stroke or carotid artery stenosis which often cause arterial transit delay errors on ASL.55, 99, 100 There are mixed findings in dementia on whether there is ATT difference between AD patients and controls.101, 102 The PLD used in ASL dementia studies is 1.5–2 s,11, 40-44, 103 while 2–3 s for stroke patients.104 ASL with multiple PLDs or multiple inflow times allows more reliable ATT estimation and PLD optimization, which may improve CBF quantification and increase the diagnostic value of ASL. In addition, velocity selective ASL is less affected by ATT and can minimize transit delay errors, and it has been shown that velocity selective ASL could identify regions with long arterial delays accurately in patients with steno-occlusive cerebrovascular disease.105 Therefore, ASL with new techniques such as velocity selective ASL may overcome the limitations of conventional ASL. Second, ASL is sensitive to head motion. ASL CBF imaging is obtained from the difference between the labeled and nonlabeled images and thus head motion (that cannot be canceled out) leads to artifact on ASL CBF image. In a recent study, Mutke et al found that the ASL images in 11 of 28 (39.3%) patients were uninterpretable due to head motion.51 Since commercial product ASL sequences (such as QUASAR-PASL and FAIR-PASL) are especially vulnerable to motion artifacts and only these commercial product ASL sequences can be used for clinical imaging, motion correction programs need to be made available in the clinical setting.50, 51, 67 In addition, paralleling imaging techniques may provide faster ASL scans with reduced motion sensitivity, and better ASL images with high sensitivity and resolution.4 Third, there are other factors that cause artifacts on ASL images or errors in CBF measurement. For example, an overestimation of ASL in deep cortical regions has been observed, which leads to higher ASL-CBF values than those measured by 15O-water-PET.30, 31 This can be corrected by using vascular crushing gradients (30). Another example is the temporal bone artifact. Musiek et al reported that radiologists could not examine the inferior temporal lobes fully on ASL images which might be due to the low image quality in the temporal bone area.43 Brain regions such as the inferior temporal lobes are close to air-tissue or bone-tissue interfaces and have less homogeneous magnetic fields, which causes signal loss in such regions on ASL.6 However, this artifact has little impact on the diagnosis of dementia.43 In addition, the differences between commercial ASL implementations from major MRI vendors may affect CBF measurements and the reproducibility of ASL (Mutsaerts et al).106 Thus, ASL parameters or options (e.g., for labeling and readout) in different commercial product sequences need to be standardized to make ASL clinically reliable across vendors. Some ASL parameters (or options) such as resolution, scan time, and vessel suppressing gradients (or vascular crusher gradients) need to be optimized for clinical use (e.g., optimize the trade-off between signal-to-noise ratio, sensitivity, specificity, and time).104 Optimization and standardization of ASL sequences, parameters, and CBF quantification methods are also important for large-scale or multi-site neuroimaging studies (that include ASL imaging) such as the Alzheimer's Disease Neuroimaging Initiative (ADNI) and the GENetic Frontotemporal dementia Initiative (GENFI), or clinical trials such as Perfusion imaging by ASL for clinical use in stroke (PEGASUS).51 In recent years, much efforts have been made to standardize ASL imaging and CBF quantification methods for clinical applications,6, 24 and more collaborations between ASL research communities and MRI vendors are still needed. In addition, since time is very limited in certain clinical applications such as stroke, it is necessary to not only make ASL scans fast, but also make ASL postprocessing fast, convenient (e.g., add lesion volume measurements) and robust to meet clinical demands. In summary, optimization and standardization of ASL parameters (and/or options) is needed to overcome the current limitations of ASL before ASL to be accepted as a clinical imaging test. Furthermore, new ASL techniques such as velocity selective ASL can minimize transit delay errors of conventional ASL and hold the promise that ASL may become an alternative to FDG-PET/SPECT/DSC in the clinical setting. Apart from the limitations of ASL, there are some methodological issues in the comparative studies covered in this review. One limitation in most of the studies is the relatively small sample sizes. The diagnostic value of ASL needs to be further validated in larger studies. Another limitation is that most comparative studies acquired ASL and FDG-PET/15O-water-PET/SPECT images sequentially over time which could introduce temporal variations in regional cerebral perfusion or metabolism. Several recent studies acquired the images concurrently31, 32, 35, 42, 43, 97 to minimize temporal variations. Direct comparison of ASL with FDG-PET/15O-water-PET/SPECT/DSC-MRI allows objective assessment of the diagnostic value of ASL, and more comparative studies with simultaneously acquired ASL and clinical images are needed to further validate the diagnostic value of ASL. In addition, there are mixed findings in the CBF measurements using ASL and 15O-water PET, and further studies are needed to make ASL CBF measurement reliable for clinical use. In conclusion, the comparative studies reviewed in this paper have demonstrated that ASL and FDG-PET/15O-water-PET had overall concordance and regional variability in healthy subjects, and regional hypoperfusion in AD and other neurological disorders identified by ASL was largely concordant with the hypometabolism/hypoperfusion seen on FDG-PET/SPECT, suggesting that ASL may have a similar diagnostic value as FDG-PET/SPECT. In addition, ASL may be less sensitive than DSC or DCE in the diagnosis of neurological disorders such as stroke. Further studies with large samples are needed to validate the diagnostic value of ASL, and its CBF quantification. For ASL to become a clinical imaging test (an alternative to invasive FDG-PET/SPECT/DSC/DCE), it is necessary to optimize and standardize ASL parameters, and explore new ASL techniques (such as velocity selective ASL) to overcome the current limitations of ASL and increase its diagnostic value in AD and other neurological disorders. This work was partially supported by the Schulich School of Medicine & Dentistry Research Initiative Grant (University of Western Ontario). The author thanks the authors and publishers who contributed their figures to this review and granted permissions for reusing their figures for illustration purpose in this paper. In particular, Figures 1 and 4 are from the Wiley publication, and at the publisher's request, the copyright notice of the figures is acknowledged here: “Permission is hereby granted for the use requested subject to the usual acknowledgements”.