Comparison of Arterial Spin Labeling and Bolus Perfusion-Weighted Imaging for Detecting Mismatch in Acute Stroke

Abstract

PURPOSE

The perfusion-weighted imaging (PWI) – diffusion-weighted imaging (DWI) mismatch paradigm is widely used in stroke imaging studies. Arterial spin labeling (ASL) is an alternative perfusion method that does not require contrast. This study compares the agreement of ASL-DWI and PWI-DWI mismatch classification in stroke patients.

MATERIALS AND METHODS

This was a retrospective study drawn from all 1.5T MRI studies performed in 2010 at a single institution. Inclusion criteria were: symptom onset<5 days, DWI lesion>10 ml, acquisition of both PWI and ASL. DWI and PWI-Tmax>6 sec lesion volumes were determined using automated software. Patients were classified into reperfused, matched, or mismatch groups. Two radiologists classified ASL-DWI qualitatively into the same categories, blinded to DWI-PWI. Agreement between both individual readers and methods was assessed.

RESULTS

51 studies met the inclusion criteria. Seven cases were excluded (1 due to PWI susceptibility artifact, 2 due to motion, and 4 due to severe ASL borderzone sign), resulting in 44 studies for comparison. Inter-rater agreement for ASL–DWI mismatch status was high (κ =0.92, 95% CI 0.80–1.00). ASL-DWI and PWI-DWI mismatch categories agreed in 25/44 cases (57%). In the 16 of 19 discrepant cases (84%), ASL overestimated the PWI lesion size. In 34/44 cases (77%), they agreed regarding the presence of mismatch versus no mismatch.

CONCLUSION

Mismatch classification based on ASL and PWI agree frequently but not perfectly. ASL tends to overestimate the PWI-Tmax lesion volume. Improved ASL methodologies and/or higher field strength are necessary before ASL can be recommended for routine use in acute stroke.

INTRODUCTION

Stroke is the second leading cause of death worldwide and the leading cause of adult disability in the United States and Europe. Routine MRI accompanied by DWI is highly sensitive in diagnosing stroke location and volume; but it only assesses the irreversibly damaged cells at the time of imaging; for this reason, information about the perfusion status of the tissue is valuable to determine the best treatment strategy. Several recent clinical trials have suggested that patients with a mismatch between their DWI lesion volume and the volume of hypoperfused tissue respond to reperfusion therapies^1–3. These studies have relied on dynamic susceptibility contrast (also known as bolus perfusion-weighted imaging [PWI]) to evaluate cerebral perfusion. Unfortunately PWI requires intravenous gadolinium-based contrast agents, which are contraindicated in patients with poor renal function or prior allergic reaction.

An alternative methodology, arterial spin labeling (ASL), can be used to acquire perfusion images⁴. ASL detects perfusion without the use of exogenous contrast, instead relying on magnetic labeling of arterial blood. Prior studies have shown that ASL can detect cerebral blood flow (CBF) alterations in the setting of acute stroke^5–13. In this study, we assessed how well ASL performs at classifying patients into clinically relevant subgroups (such as reperfusion, matched lesion, or mismatched lesion), using a quantitative DWI-PWI mismatch methodology as a reference standard.

MATERIALS AND METHODS

Patient Population

This retrospective study was approved by our Institutional Review Board and was HIPAA compliant. All inpatient MRI brain studies performed in our institution during a single calendar year (2010) were evaluated. Inclusion criteria were as follows: 1) interval between the onset of the symptoms and the MRI study <5 days; 2) DWI, PWI, and ASL obtained; 3) DWI lesion volume >10 ml.

MR Imaging

Imaging was performed at 1.5 T (Signa; GE Medical Systems, Milwaukee, WI, USA). All patients received axial DWI (TR/TE 6000/70 ms, b=1000 sec/mm2 isotropic, 5 mm section thickness skip 1.5 mm) and an MR angiogram covering the circle of Willis (3D TOF, TR/TE 30/2 ms, 1.5 mm section thickness skip 0 mm). For bolus PWI, gradient-echo planar imaging (EPI) was performed during the passage of 0.1 mmol/kg of gadopentetate dimeglumine (Berlex Laboratories, Wayne, NJ) or gadodiamide (GE Healthcare, Waukesha, WI), administered at a rate of 4 ml/sec. Image readout was performed using single-shot EPI (TR/TE 2000/60 ms) or multi-shot multi-echo generalized auto calibrating partially parallel acquisition (GRAPPA) EPI (TR/TE 1225/[17, 30, 52] ms); acceleration factor of 3, 5-mm section thickness skip 1.5 mm). The in-plane spatial resolution was 2.6 mm. Acquisition time was 2 min. ASL was performed using pulsed continuous labeling¹⁴ with the following parameters: TR/label time/post-label delay [PLD]/TE 5500/1500/2000/2.5 ms, three-dimensional background-suppressed fast-spin-echo stack-of-spirals readout, 4 mm in-plane and 6 mm through-plane resolution, 6 min acquisition time. Vessel suppression was not performed. An automated reconstruction script returned CBF images directly to the scanner console within 1 min.

Data Analysis

Large artery steno-occlusive disease was defined as >70% narrowing or occlusion of one of the major arterial vessels, which included the internal carotid artery, middle cerebral artery (M1 segment), posterior cerebral artery, basilar artery, or vertebral arteries) as seen on the MR angiogram. DWI and PWI were evaluated using automated mismatch software called RAPID¹⁵, which calculates DWI and PWI (Tmax>=6s) volumes, and was the reference standard to determine the mismatch status. To this end, we used the modified DEFUSE criteria¹⁶ to divide patients into one of three classes: (1) “reperfused” (PWI volume<70% of DWI volume, and absolute reperfusion volume>10 ml); (2) “matched” (PWI volume>70% but <180% of DWI volume); and (3) “mismatched” (PWI volume>180% of DWI volume, and absolute mismatch volume> 10 ml). To determine ASL-DWI mismatch status, two radiologists evaluated the DWI and ASL images in a qualitative manner, blinded to the PWI images and DWI-PWI mismatch result. The primary analysis method (Method #1) was to consider only the regions with low ASL signal as the perfusion abnormality; a secondary assessment was also performed in which the “ASL lesion” included both the low ASL signal regions as well as surrounding regions with serpiginous high signal that were thought to represent regions of delayed arterial arrival (Method #2).

Statistical Analysis

Kappa statistics (both unweighted and weighted) were calculated to assess agreement between readers for ASL-DWI classification, as well as the agreement between the mismatch status based on PWI-DWI and ASL-DWI. We further grouped the patients based on presence or absence of mismatch: in this case, those patients with either reperfusion or matched status were combined and compared with those with a mismatch status. Finally, we tested whether there was a difference in agreement based on the presence of large artery steno-occlusive disease.

RESULTS

Fifty-one patients met the inclusion criteria. Seven exams (13.7%) were subsequently deemed uninterpretable: 1 study with PWI susceptibility artifact (1.9%), 2 studies with motion artifacts (3.9%) and 4 studies with severe ASL borderzone sign (7.8%)¹⁷. This left 44 studies in 43 patients (mean age 59±16 years, range 15–85 years; 22 men/21 women) with mean DWI lesion volume of 66±56 ml (range 10–243 ml) and bolus PWI lesion volume (Tmax>=6 sec) of 64±64 ml (range 0–219 ml). Time from last seen normal was 37±24 hrs (range 2–111 hrs). The indications were as follows: acute hemiparesis with or without aphasia (n=22), altered mental status (n=5), post-surgery infarct (n=5), vasospasm (n=4), isolated aphasia (n=2), unresponsiveness (n=2), visual field cut (n=2), extensor posturing (n=1), and sensory disturbance (n=1). Twenty-six patients (59%) had evidence of large artery steno-occlusive disease.

The agreement between two readers for ASL-DWI mismatch status (Method #1) was almost perfect, with agreement in 42 of 44 cases (95%), yielding unweighted and weighted kappa of 0.92 (95% confidence interval [CI]: 0.80–1.00) and 0.94 (95% CI: 0.75–1.00), respectively (Table 1). This agreement was better than for Method #2 (see Supplemental Table 1).

Table 1.

Agreement between the two readers for ASL-DWI mismatch status (Method #1).

	Reader 1
Reader 2	Reperfusion	Matched	Mismatched	Total
Reperfusion	5	1	0	6
Matched	1	25	0	26
Mismatched	0	0	12	12
Total	6	26	12	44

Unweighted kappa: 0.92 (95% CI: 0.80–1.00)

Weighted kappa: 0.94 (95% CI: 0.75–1.00)

Based on PWI-DWI, 19 patients had reperfusion (43%), 17 patients had matched lesions (39%), and 8 patients had a mismatch (18%). Agreement between the ASL-DWI mismatch status and reference standard PWI-DWI classification was moderate, and is described in Table 2 (for Method #1) and in Supplemental Table 2 (for Method #2). For Method #1, the two approaches agreed in 25 of 44 cases (57%) with an unweighted kappa of 0.35 (95% CI: 0.13–0.57) and a weighted kappa of 0.42 (95% CI: 0.12–0.73). In 16 of the 19 cases with discrepancy (84%), it was because the perfusion lesion was deemed to be larger on ASL than on PWI. There was no significant difference in kappa values between PWI-DWI and ASL-DWI classification based on the presence of large artery steno-occlusive disease (unweighted kappa: large artery steno-occlusive disease 0.35 [95% CI 0.05–0.64] versus none 0.29 [95% CI 0.00–0.66]; weighted kappa: large artery steno-occlusive disease 0.36 [95% CI 0.11–0.60] versus none 0.34 [95% CI 0.08–0.60]).

Table 2.

Agreement between ASL-DWI (Method #1) and PWI-DWI mismatch classifications

	PWI-DWI Classification
ASL-DWI Classification	Reperfusion	Matched	Mismatched	Total
Reperfusion	7	0	0	7
Matched	9	13	3	25
Mismatched	3	4	5	12
Total	19	17	8	44

Unweighted kappa: 0.35 (95% CI: 0.13–0.57)

Weighted kappa: 0.42 (95% CI: 0.12–0.73)

When comparing only mismatch versus no mismatch (i.e., no mismatch and reperfusion cases combined), agreement increased to 34/44 cases (77%). ASL-DWI had the following test statistics for identifying mismatch patients: sensitivity 0.63 (95% CI 0.26–0.89), specificity 0.81 (0.63–0.91), positive predictive value 0.42 (0.16–0.71), and negative predictive value 0.91 (0.73–0.98). For Method #2, the results were similar, but with slightly lower rates of agreement and kappa values (Supplemental Table 2).

Figures 1–3 demonstrate examples of reperfusion, matched, and mismatched patients in which there was agreement between the two methodologies, respectively. Figure 4 shows a case of disagreement; the ASL-DWI status was thought to represent a mismatch, while the PWI-DWI analysis classified the case as reperfusion. In this case, it is clear that a large fraction of the ASL perfusion lesion corresponds with a region of prolonged Tmax on PWI, but the Tmax in this region was too mild to be included in the mismatch calculation (i.e., <6 sec).

Figure 1.

57 year-old woman two days after onset of aphasia demonstrates a DWI lesion in the left MCA territory (a). (b) PWI-Tmax and (c) ASL demonstrate reperfusion of the lesion, and accordingly this case was scored as “reperfusion” using both PWI-DWI and ASL-DWI approaches. Hyperemia is clearly visualized on ASL images, compatible with luxury perfusion.

Figure 3.

53 year-old woman with right-sided weakness and aphasia, imaged 21 hrs following onset of symptoms. (a) DWI shows a moderately sized lesion in the left MCA territory. (b) A larger volume of PWI Tmax greater than 6 sec is present, compatible with a mismatch. (c) ASL also demonstrates a larger volume of low signal, with surrounding serpiginous high signal compatible with slow flow in collaterals. Based only on the region of low ASL signal, this case was scored as a mismatch, consistent with the PWI-DWI method.

Figure 4.

86 year-old woman status post attempted aneurysm coiling, with left-sided deficits. (a) DWI demonstrates scattered regions of acute infarction, with a total lesion volume of 39 cc. (b) PWI Tmax maps demonstrate a large region of relatively mild abnormality. Only 19 cc of this lesion were deemed to have a Tmax of greater than 6 sec, and for this reason represents a case of reperfusion. (c) ASL demonstrates reduced signal in the entire PWI lesion, encompassing all of the mild Tmax prolongation, and based on this was scored as a mismatch. The increased sensitivity of ASL in region with mildly prolonged arterial arrival was a common cause of discrepancy between the two techniques.

DISCUSSION

In this report, we compare non-contrast ASL with contrast PWI for classifying patients into clinical categories relevant for mismatch-based acute stroke studies. Some patients cannot receive MR contrast agents, so it is important to know whether similar information can be obtained using a non-contrast method, such as ASL. Furthermore, there is some suggestion that ASL might be particularly sensitive to evaluate collateral flow^{18, 19}, which is itself relevant in acute stroke^{20, 21}.

We found that diagnostic quality ASL images can be obtained in stroke patients even at low field strength (1.5T) and that two readers can agree on mismatch categories. Furthermore, there was moderate agreement between the two methods, and this agreement increased when the categories were collapsed into mismatch versus no mismatch. In particular, we wish to highlight two observations: (1) when ASL showed either a matched or reperfusion pattern, it was almost always corroborated by the PWI-DWI methodology (i.e., negative ASL had a high negative predictive value for excluding a mismatch); and (2) when ASL showed a mismatch, it was essentially a toss-up as to whether a PWI-DWI mismatch existed (i.e., mismatch on ASL-DWI had a low positive predictive value for PWI-DWI mismatch). Both points stem from the fact that ASL as currently employed is “over-sensitive” to mild perfusion deficits and arterial delays that do not reach the PWI Tmax>6 sec criteria used in clinical trials. This finding is similar to those reported by Siewert et al., in which qualitative agreement between a pulsed ASL technique (EPISTAR) and PWI was present in 17 of 21 patients, with the discrepant cases representing absent perfusion on ASL in patients with delayed perfusion based on PWI⁵. These findings are not dissimilar to the initial reports of the use of PWI in acute stroke, where it was shown that mild mean transit time (MTT) and Tmax prolongations included regions of benign oligemia that did not infarct even in the absence of reperfusion²², and which led to the development of new criteria.

It also suggests a solution to the problem: if ASL is to be used for this application, it must be made less sensitive to mild arterial delays. This can be accomplished using either longer post-label delay times²³ or the use of an arterial transit delay insensitive technique such as velocity selective ASL (VSASL)²⁴. Unfortunately, both of these methods have lower SNR than the ASL sequences typically used for clinical use. Use of higher field strength, such as 3T and 7T, will likely enable more routine use of these sequences in clinical patients. Alternatively, such specialized ASL methods could be reserved for patients referred for acute ischemic stroke rather than as a “one-size-fits-all” approach. The fact that we had to exclude 8% of the patients from this study who had global severely prolonged arterial arrival time – evidenced by a severe ASL borderzone sign¹⁷ – also indicates the need for improved arrival time independent ASL methodologies.

The ASL sequence employed in this study uses pulsed continuous labeling, background suppression, and fast spin-echo readout, leading to high SNR and reduced motion sensitivity¹⁴. It did not, however, employ diffusion crusher pulses to reduce the conspicuity of flowing spins within large vessels, a technique known as vessel (or vascular) suppression^25,26. For this reason, we estimated the volume of abnormal ASL based on two criteria, meant to simulate images that would be acquired with and without vessel suppression (Methods 1 and 2, respectively). We found better inter-reader and inter-modality agreement for Method 1, in which we considered only the low ASL region to represent low perfusion, akin to vessel suppressed ASL. We found that it was possible to agree on the size of the hypoperfused region even in the presence of significant arterial transit artifact (ATA). For routine clinical indications, we have found that the ability to depict slow flow in collateral networks using non-vessel-suppressed ASL outweighs the potential confounding effects of ATA. One caveat is that reperfusion and slow flow can potentially be mistaken for one another, given that they both show high ASL signal; in practice, we did not find this distinction to be overly challenging, as shown by the different character of the high signal between the two states (see Figures 1 and 3, for example). However, it is probably that a vessel-suppressed ASL technique would work as well or even possibly better in the acute stroke setting for detecting mismatch.

One major limitation of ASL, as for any CBF technique, will be determining lesion volumes using a threshold method, as is currently done in many stroke trials for DWI and PWI Tmax measures. In this study, we performed a qualitative assessment, and acknowledge the inherent problems with this approach (such as reproducibility) that would limit its applicability for a large-scale imaging stroke trial. The problem stems from the marked differences between normal gray and white matter CBF, as well as the overall low CBF values present in normal white matter. Approaches to this are being developed within the CT perfusion literature, since currently cerebral blood volume and CBF are thought to give the best approximations of the stroke core region²⁷, and include automated approaches that compare ipsilateral to contralateral regions.

There are several limitations to our study. The first is the retrospective nature of the study, which can introduce unknown biases. Also, for this reason we were unable to establish NIHSS scores on all patients. The second is the relatively small number of cases; this is largely because ASL is still not widely used in acute stroke. In our institution, this is dictated by the need for rapid patient triage. The mean time to imaging in this study (37 hrs) is relatively late, making it difficult to relate the findings to potential changes in treatment. While the late time frame had the advantage of yielding a good distribution of mismatch statuses, we believe that the findings in this study need to be validated in a larger, prospective cohort of hyperacute stroke patients. An ideal study would furthermore be done using a longer PLD conventional ASL or VSASL on high-field scanners. Finally, while the Tmax>6 sec threshold appears to be a promising cutoff value for identifying penumbral tissue, it is important to note that other perfusion parameters, such as the mean transit time or non-deconvolution-based time-to-peak (TTP)²⁸, may yield different results.

Lastly, we point out that some of the discrepancy between the two techniques may arise because they are examining different things (CBF with ASL vs. arterial delay with PWI Tmax). Several groups have endeavored to create arterial arrival delay maps using multi-delay ASL, and these may yield better correspondence with PWI Tmax maps^{10, 29–31}. However, these methods tend to require longer acquisition times and have lower SNR than ASL CBF maps. While PWI Tmax has become a reference standard for acute stroke studies and prior work has suggested that there is an inverse relationship between Tmax and CBF in acute stroke^{32, 33}, any future studies should also examine the relationships between CBF and patient outcomes.

CONCLUSION

There is good inter-reader agreement using a qualitative estimation of the perfusion-DWI mismatch using ASL as the perfusion method. We found better agreement between readers as well as with a reference standard (PWI-DWI) if the ASL perfusion deficit is defined as the region of low signal, disregarding serpiginous high ASL signal that represents slow flow in collateral vessels. Using this approach, ASL and PWI agree a little over half of the time regarding mismatch status, and this agreement rises to 77% for the distinction between mismatch and no mismatch. In almost all discrepant cases, ASL overestimated the size of the PWI-Tmax>6 sec lesion. This is due to ASL’s high sensitivity to mild arterial arrival time delay. Improved ASL methodologies (such as VSASL or long PLD ASL) and use of higher field strength are likely necessary before ASL can be recommended for routine use in acute stroke.

Figure 2.

48 year-old woman imaged 26 hrs following onset of a new right visual field cut. (a) A small DWI lesion is noted in the left PCA territory. (b) PWI Tmax and (c) ASL show a small focus of reduced perfusion in the same region. Both PWI-DWI and ASL-DWI approaches classified this lesion as a matched defect.