Neuroimaging in acute ischaemic stroke: insights into unanswered questions of pathophysiology

Abstract

Wardlaw JM (Western General Hospital, Edinburgh, UK). Neuroimaging in acute ischaemic stroke: insights into unanswered questions of pathophysiology (Review). J Intern Med 2010; 267: 172–190.

Abstract.  The treatment of acute ischaemic stroke is based on the principle that there is ischaemic but still potentially salvageable tissue that could be rescued if blood flow could be restored quickly. It is assumed that salvage might only be possible in the first few hours, and that infarct expansion is a direct result of failed recanalization of the main artery. This concept arose from experimental work in the 1970s, supported more recently by studies using imaging to identify penumbral tissue. However, although magnetic resonance diffusion and perfusion imaging is a way of imaging penumbral tissue and has been around for over a decade, it is not an easy technique to apply in practice and its use has produced conflicting results. Computed tomography perfusion, and any other tissue perfusion imaging technique, is likely to encounter the same difficulties. Indeed many factors, other than the presence of a diffusion-perfusion mismatch acutely, may determine or influence ultimate tissue fate even days after the stroke, and in turn, clinical outcome. Many of these potential influences are beginning to emerge from studies using different forms of imaging at later times after stroke. This review will explore the information now emerging from imaging studies in large artery ischaemic stroke to summarize knowledge to date and indicate unresolved issues for the future.

Introduction

The traditional view of the ischaemic lesion

This review will concentrate on large artery (or cortical) ischaemic stroke, about which the most is known concerning pathophysiological mechanisms. Most strokes are ischaemic in nature and due to sudden occlusion of a cerebral artery, leading to sudden reduction in blood flow in the territory of the occluded artery. This has led to a major focus of stroke research and treatment being on the first few hours after stroke, neglecting events that occur at later, subacute, stages and that may in themselves significantly and adversely affect the final outcome. Final clinical (and radiological) outcome is the sum total of all preceding events, not just the first few hours. Whilst the first few hours are undoubtedly critical to the damage caused by the stroke, neglect of important events occurring subacutely means that we know little of the natural history of the subacute phase of ischaemic stroke and are probably missing opportunities to intervene to prevent additional cumulative damage.

The original and elegant experimental work of Astrup et al. [1] suggested that middle cerebral artery (MCA) occlusion resulted in death of neurones that was both flow and time dependent – the worse the drop in flow and the longer it went on, the greater the volume of dead tissue. However they also demonstrated that less severe reductions in flow were compatible with survival of neurones in a shut down but still viable state, the penumbra [2]. In other words, as blood flow fell, the electrical activity ceased at a level where cell membranes and most other cellular functions were still intact [2]; as the blood flow fell further, the mechanisms to preserve basic cellular integrity, such as cell membrane ion pumps, would also start to fail resulting in progressive cellular morphological changes [3] that would eventually reach the point of nonviability, even if flow were then restored [4, 5].

The blood flow levels at which these key events – loss of neuronal electrical activity which the patient would notice as the stroke symptoms, followed by loss of cellular integrity with cell swelling and eventual death – appeared to be fairly consistent across all laboratory animal species that were tested (Fig. 1) [6, 7] resulting in the assumption that there were areas of the lesion where flow was uniformly reduced below these key levels and an expectation that similar thresholds would apply in humans. Indeed, early patient studies, using positron emission tomography (PET) imaging, did seem to find broadly similar tissue levels [8–10], albeit (because of the limited availability and complexity of these techniques) in studies with few subjects. These studies, and the apparent consistency of the blood flow levels across all mammals tested, gave considerable encouragement to the idea that it would be possible to measure blood flow levels in patients using imaging techniques, thereby to determine the volume of penumbral and dead tissue, and use this information both to test new therapies and to guide decisions on use of established treatments – no salvageable tissue, no treatment (Table 1).

Figure 1.  Critical blood flow levels for cell death and tissue at risk of infarction (penumbra) derived from experimental models and corresponding information from imaging studies [1].

Table 1.   Characteristics, advantages and limitations of imaging used in stroke

Modality	What does it show?	Availability	Pros	Cons
CT, computed tomography; MR, magnetic resonance; eGFR, estimated glomerular filtration rate (mL 100 g^−1 min^−1).
Positron emission tomography	Cerebral blood flow; cerebral oxygen uptake and glucose usage	Very limited; radiotracers have very short t½; requires intra-arterial line	Provides fully quantitative measure of blood flow and cerebral metabolism; can distinguish infarct dead core from penumbra and benign oligaemia	Limited availability; not suitable for rapid acute diagnosis; requires intra-arterial access limiting concurrent thrombolysis treatment
Plain CT	Acute and old ischaemia; tissue density change proportional to change in water content of tissue, i.e. to amount of oedema in acute stroke	Very widespread	Rapid exclusion of haemorrhage and stroke mimics; provides diagnostic quality images in virtually all acute stroke; early infarct signs visible in approx. 50% (more amongst severe stroke) and specific for ischaemic stroke	Early infarct signs difficult to recognize (dark on dark background); ischaemia not visible in large proportion of patients with mild stroke, so lacks positive diagnosis
MR structural (T2, FLAIR, T1, T2*) and diffusion imaging	Structural: Acute and old ischaemia – signal change proportionate to change in water content of tissues, i.e. to amount of oedema. Diffusion – very sensitive to acute ischaemia	Equipment in many hospitals but limited access for stroke	T2, T1 and FLAIR of similar sensitivity as CT to acute ischaemia; T2* very sensitive to old haemorrhage; acute haemorrhage can be difficult to recognize easily; diffusion highly sensitive to acute ischaemic lesions; positive diagnosis of ischaemia even very early after stroke; lesion easily seen as white on dark background	T2, T1, FLAIR sensitivity to acute ischaemic no better than CT; must do diffusion imaging; less available than CT; scanning takes longer; cannot be used in patients with contraindications to MR and not well tolerated in severe acute stroke
CT perfusion imaging	Semiquantitative and relative maps of cerebral blood flow, blood volume and mean transit time	Potentially available on most CT scanners where appropriate software added	Rapid assessment of perfusion deficit; may identify tissue at risk but more data required	Substantial radiation dose; requires i.v. contrast injection; avoid in patients taking metformin, with renal failure or allergies; as yet limited data on accuracy of perfusion maps or predictive ability
MR perfusion imaging	Semiquantitative and relative maps of cerebral blood flow, blood volume and mean transit time	As per MR scanners, where software is available	Rapid assessment of perfusion deficit; may identify tissue at risk but more data required	Requires i.v. contrast injection; contraindicated in renal impairment (eGFR <60 mL 100 g^−1 min^−1 relative; <30 mL 100 g^−1 min^−1 absolute); no consensus on which perfusion parameter should be used to quantify tissue at risk
CT angiography	Stenosis or occlusion in arteries and veins	On most CT scanners	Rapid assessment of stenosis or occlusion of artery	Requires IV contrast injection; shows proximal major intracranial arteries; substantial radiation dose; contraindications as above
MR angiography	Stenosis or occlusion in arteries and veins or dissection	On most MR scanners	Rapid assessment of stenosis or occlusion of artery; thrombus in wall of artery in dissection	Requires i.v. contrast injection; shows proximal major intracranial arteries; contraindications to MR and to contrast as above

It is more than 30 years since the original experiments that led to these concepts. There are now several ways of imaging penumbral tissue, which have been used in clinical trials to select patients and monitor treatment [11–14]. But perfusion thresholds are not used in routine clinical practice, and there has been little success in translating the apparent therapeutic benefits of new treatments from positive studies in animal models to either positive results in clinical trials or to real benefits for patients. Whilst this latter failure may reflect many factors, such as the heterogeneity of stroke in human subjects compared to model systems, the shortcomings in methodologies used in many animal studies [15], and differences in physiology and pharmacology between humans and other mammals, it is also likely that there are other explanations. These could include application of complex imaging techniques and limited appreciation of the full complexity of ischaemic stroke in humans [16–20].

The one point of which we can be certain however, and which provides strong support for the existence of a salvageable penumbra even if there have been problems translating the knowledge into greater treatment precision, is the now overwhelming evidence that restoring blood flow to the ischaemic tissue by removing the occlusion, either spontaneously or therapeutically, leads to better recovery from the stroke compared with persistent occlusion of the artery [21, 22]. There are several pieces of evidence for this. First, a systematic review of all available observational studies (about 1000 patients) demonstrated a sixfold increase in the odds of good functional outcome at 3 months if recanalization (monitored by a range of methods) occurred within 5 h of stroke, and a fourfold increase in the odds of good functional outcome at 3 months where recanalization occurred within 24 h of stroke [21]. Secondly, thrombolytic treatment, which increases both the amount and speed of recanalization, improves functional outcome [22]. Although individual randomized trials had mixed results, the sum total of all the thrombolysis trial evidence (26 trials, about 7000 patients) indicates that thrombolysis given up to 6 h, possibly 9 h or even later, reduces the risk of poor functional outcome (odds ratio 0.81, 95% confidence interval 0.73, 090) [22]. Whilst faster recanalization of the occluded artery is assumed to be the main treatment mechanism of thrombolysis, several studies demonstrated that in patients with a hyperattenuated artery (as a surrogate for an angiographically occluded artery) at presentation, subsequent disappearance of the hyperattenuated artery was associated with less infarct swelling (Fig. 2) and better functional outcome than in patients in whom the hyperattenuated artery persisted [23, 24].

Figure 2.  Example of CT scan from a patient with an acute ischaemic stroke affecting the left middle cerebral artery (MCA) territory. Top row, at 4 h after stroke onset there is a hyperattenuated left MCA main stem (arrow) indicating acute occlusion with thrombus and some early low attenuation with loss of grey-white matter differentiation in the left basal ganglia, frontal, temporal and parietal lobes in keeping with early ischaemic change. Bottom row, at 48 h, there is persistent left MCA main stem occlusion (hyperattenuated artery, arrows) and now a massively swollen infarct involving most of the left MCA territory. Patients with persistent hyperattenuated artery (occlusion) have worse clinical and radiological outcomes and more infarct swelling [21, 23, 24].

The remainder of this review will explore current knowledge on the most available imaging techniques for assessing tissue injury in stroke, not just in the first few hours but right throughout the subacute phase. It will also highlight new thinking on aspects of human ischaemic stroke pathophysiology hitherto unrecognized or overlooked, that may offer new opportunities for stroke treatment to optimize recovery after large artery ischaemic stroke.

Methods for imaging the ischaemic lesion

Positron emission tomography imaging was the starting point for showing that penumbral, and therefore potentially salvageable, tissue did indeed exist in patients for prolonged periods of time [9, 10, 25], perhaps as long as 24 or 48 h after the stroke [25, 26]. However, PET is a complex and not widely available imaging technique.

Magnetic resonance (MR) imaging for patients emerged in the 1980s, but the basic T2 and T1-weighted structural sequences are really no more sensitive than computed tomography (CT) for demonstrating acute ischaemic lesions, they are less specific than CT, less practical, take longer and miss other causes of sudden focal neurological deficits (like meningiomas). So MR did not really excite much interest for stroke initially. It was in the 1990s, with the development of diffusion imaging and perfusion imaging several years later, that the major potential for application in stroke became apparent. Diffusion imaging is very sensitive to acute ischaemia, becoming abnormal within minutes of large artery occlusion [27, 28]. The ischaemic lesion is very easy to see by virtue of being bright white (or hyperintense) on a dark background – the so-called ‘light bulb sign’– in contrast to CT scanning, where the acute lesions are slightly darker than an already dark background. Thus the acute ischaemic lesions on diffusion imaging were easy to see even for those with no specialist training.

Diffusion imaging mainly demonstrates the mobility of water molecules in the extracellular space. Anything that restricts extracellular water molecule movement, such as even mild cell swelling, will restrict water movement and show up on diffusion imaging as hyperintense. The restriction in water movement is expressed in the apparent diffusion coefficient (ADC) image, the reduced water movement showing up as dark. ADC maps also show some underlying brain structure, e.g. major white matter tracts, as they are sensitive to any restriction of water movement in particular directions resulting from normal brain anatomy, which is the basis of using diffusion imaging to map white matter tracts.

Perfusion imaging was introduced in the mid to late 1990s as faster scanners became available [29]. It can be performed with a contrast agent given intravenously (so called ‘susceptibility-weighted imaging’) [30, 31], or use the inherent properties of flowing blood entering the imaging slice (arterial spin labelling) [32] to image cerebral blood flow (CBF). Susceptibility-weighted perfusion imaging is fast and more practical than arterial spin labelling in stroke patients and so most information on MR perfusion imaging in stroke is based on the use of intravenous contrast bolus tracking methods. The problem with arterial spin labelling has been that the signal to noise ratio is very low and patients have to keep very still during very long scan acquisition times, making it so far a relatively impractical technique for use in acute stroke.

Computed tomography perfusion imaging is now also available [33–35], in which the patient is scanned rapidly through the brain during the first pass of an intravenously injected bolus of contrast agent. Studies so far suggest that both CT/CT perfusion (CTP) and MR diffusion weighted imaging (DWI)/perfusion weighted imaging (PWI) may yield similar information in the acute phase in comparable patients [36, 37]. As individual centres have differing experience of their use, and different access, future trial recruitment will likely be optimal (and the results more generalizable) if both CT and MR perfusion are sufficiently well understood so that the results are interchangeable. The practical limitations of MR, particularly in patients with more severe acute stroke symptoms, should not be underestimated [38–40]. MR is not easy to perform in hyper-acute stroke [38–40], particularly when combined with the need to administer thrombolysis. As an indication of the difficulties, the DIAS [11], DEDAS [12], DEFUSE [41] and EPITHET [14] studies of thrombolysis in patients who had MR DWI/PWI imaging each took several years to recruit 104, 37, 74 and 100 patients respectively in multiple, well-resourced centres. The rising popularity of CT/CTP simply reflects the much greater availability of CT, and its rapidity and ease of access. Future acute stroke treatment trials should consider both MR and CT in their design in order to be as efficient, cost-efficient and generalizable as possible: faster patient recruitment reaches sample targets faster, so limiting trial staff infrastructure costs; if more centres can participate, results would be more generalizable to future clinical use.

Many of the principles behind MR perfusion imaging apply to CT, with only a few important differences [42, 43]. One main difference is that the MR signal change following intravenous contrast is not linearly related to the contrast agent concentration, whereas on CT the change in attenuation is linear with contrast concentration. The other major difference is that many CT scanners so far only provide a few brain slices, sometimes as few as two, thereby giving limited brain coverage compared with MR which gives whole brain coverage. The use of intravenous MR contrast is contraindicated in patients with impaired renal function [contraindicated relatively below an estimated glomerular filtration rate (eGFR) of 60 mL min⁻¹ and absolutely below 30 mL min⁻¹]. Determining renal function may delay perfusion imaging. CT contrast is generally accepted for cautious use except in patients with renal failure (eGFR<30) [44, 45] but is contraindicated in diabetic patients on metformin. This review will focus on information provided by contrast bolus imaging and should be taken to apply to MR or CT perfusion imaging unless specified.

What does diffusion imaging tell us?

Diffusion imaging was thought initially to demonstrate permanently damaged tissue, i.e. the core of the ischaemic lesion. It was also considered that it should be possible to use specific ADC levels (i.e. an indirect measure of the amount of cell swelling) to indicate thresholds for dead versus salvageable tissue. Several ADC thresholds have been reported for apparently salvageable/nonsalvageable tissue [46–50], but these show considerable overlap between tissue that had already infarcted, that was ischaemic but viable, or that was normal but subsequently infarcted [46, 51]. Whilst it is true that lower ADC values are associated with larger, more severe strokes (Fig. 3) [52–55], different studies have quoted ADC ratios (of abnormal to normal tissue) as different as 0.8 down to 0.6 – but a threshold ratio of 0.6 would have missed many smaller and milder ischaemic lesions in our patient cohorts (Fig. 3) [52].

Figure 3.  (a) Apparent diffusion coefficient (ADC) and stroke subtype, N = 120 patients [52]; ADC and outcome. (b) Ninety-eight patients with range of stroke severities, prospectively collected, systematically and independently followed up at 3 months. Lower ADCs are associated with poor outcome but large overlap with good outcome, so not an independent outcome predictor [113].

Why the inconsistency in thresholds for identifying lesions? Diffusion lesions are rarely simple, single lesions with nice crisp edges. It is much more frequent to find patchy, irregular mixed signal lesions such as in Fig. 4a. Not only that but, whilst applying an ADC threshold to delineate a lesion might improve the repeatability of diffusion lesion volume measurement by producing a more consistent result than hand tracing, it does not improve the validity (Fig. 4b). Unfortunately, all ADC thresholds described above either underestimate or overestimate lesion volume compared with visual assessment of the visibly damaged tissue, ending up either missing out some of the definitely abnormal tissue (see Fig. 4b, ADC lesion determined using threshold 0.00055 in the centre compared with visual outline of the DWI lesion on the far left) or including normal tissue in the lesion volume (see Fig. 4b, ADC lesion determined using threshold 0.00075 on the far right compared with visual outline of the DWI lesion on the far left) [56]. Note that the pixel volumes for each lesion given on the bottom row of Fig. 4b are derived after the addition of a manually drawn outline in addition to the threshold, necessary to exclude erroneously included tissue well outside the visibly abnormal area (e.g. in the opposite hemisphere). The difference in lesion volume between these two extreme thresholds (106242.22 vs. 222011.72) represents a more than doubling of lesion volume, an unacceptable variability for clinical decision-making.

Figure 4.  Difficulty of determining diffusion lesion volume and the effect of applying automated apparent diffusion coefficient (ADC) thresholds [56]. (a) Hand tracing the margin of the hyperintense signal on a DWI lesion – look at this lesion closely and decide where you would place the outline. (b) Applying ADC thresholds does not solve the problem – too abnormal a threshold (<0.00055, central image) excludes abnormal tissue and a less abnormal threshold (<0.00075, right-hand image) includes too much normal tissue (note the white areas in the right hemisphere have been included in the abnormal tissue volume). Applying these two thresholds results in a more than twofold difference in lesion volume.

Considering the mechanism by which the tissue becomes hyperintense on diffusion imaging and what we know about the pathophysiology of acute ischaemia, it should be fairly obvious that lesions on diffusion imaging cannot possibly just represent permanently damaged tissue and that diffusion abnormalities could resolve completely (Fig. 5). Whilst ADC values in acute stroke correlate with metabolites such as lactate that indicate the presence of ischaemia, they do not correlate with markers of neuronal death such as N-acetyl aspartate (NAA) [57–60]. Intracellular oedema begins as the blood flow level falls to the point where cell membrane ion pumps begin to fail, so will start at the transition point between tissue where the cells go from being only electrically silent to becoming swollen because of membrane pump failure, i.e. truly penumbral, rather than at the point where cells go from penumbral to dead (Fig. 6). Therefore, inevitably some diffusion abnormal tissue can cover. It also means that, up to a point, diffusion imaging will become more abnormal the more the cells swell. Therefore, there should be some correlation between the degree of hyperintensity of the lesion on diffusion imaging and the amount of cell death, i.e. whiter diffusion lesions should contain more dead neurones (less NAA) than paler lesions in acute stroke, which is indeed true [60, 61]. It also means that if the blood flow is restored in time, that the cell swelling should resolve and the diffusion abnormality should disappear [46, 62, 63].

Figure 5.  Two patients illustrating different patterns of diffusion lesion evolution. In the top row, the diffusion abnormality resolved completely and the patient made an excellent recovery. The bottom row shows a different patient with a similar diffusion and perfusion lesion at presentation, but in whom the abnormal diffusion signal persisted in parts of the lesion for many weeks – in these areas blood flow failed to return to normal and the patient made a poor recovery [95]. The explanation for this and the importance of the ‘no reflow phenomenon’ is provided in the text and Fig. 9.

Figure 6.  A typical signal-time curve obtained during the first pass of the contrast bolus through the brain on MR or CT perfusion imaging. The different parameters that can be extracted from the perfusion signal time curve, and whether they represent cerebral blood flow (CBF), cerebral blood volume (CBV) or mean transit time (MTT) are illustrated.

A systematic review of diffusion imaging in stroke several years ago [64] highlighted shortcomings in the literature which precluded drawing reliable conclusions on the specificity of diffusion imaging for differentiating core from penumbra at that time. Kranz and Eastwood updated that systematic review trying to determine whether diffusion lesions represented the ischaemic lesion core or not [63]. They noted wide variability in the observed rates of diffusion lesion reversal (0–83%) with a mean rate of 24%, and concluded that the ‘evidence supporting the hypothesis that diffusion imaging is a surrogate marker for ischaemic core in humans is troublingly inconsistent’ [63].

This interpretation of what diffusion imaging abnormalities represent in stroke patients is consistent with studies in experimental models where imaging findings were compared with histology [65]. Reversible ischaemia models show that diffusion lesions disappear if perfusion is restored quickly, that the ADC level falls quickly to low levels at the onset of ischaemia and returns to normal if reperfusion is initiated quickly, and that the lesion ADC level on its own does not reflect whether the tissue can recover or not, thus confirming what we know from patient studies [65]. Interestingly, what the experimental studies show but that human studies have in general overlooked is that much of the oscillation in ADC values and diffusion lesion appearance with vessel occlusion/reopening is due to changes in glial cells not neurones. Indeed, in experimental studies, the neurones were said to become ischaemic with vessel occlusion and be less likely to recover than the glial cells with recanalization. If the experimental studies are correct, and if they mirror human ischaemic stroke, then it is predominantly glial cell swelling that produces the diffusion lesion appearance and reduces the ADC at the onset of ischaemia. It is also likely to be the glial cell changes that then either lead to resolution of the diffusion lesion or persistent infarction when the artery recanalizes. This is important because it means that much of the focus so far in designing treatments for stroke has focused on neuroprotection rather than on glial cells. Glial cells are extremely important for maintaining the neuronal milieu, so that ensuring their recovery is likely to be as important as recovering as many neurones as possible.

Tissue perfusion imaging

The other part of imaging the penumbra assumes that the lesion delineated by perfusion imaging also indicates the boundary between ischaemic but viable tissue and the oligaemic but not at-risk tissue (Fig. 1). However, here the story gets even more complicated. The problem with perfusion imaging, whether it be CT or MR contrast bolus perfusion imaging, is that there is little agreement on how the perfusion abnormality should be visualized [66–68]. The Acute Stroke Imaging Research Roadmap [69] highlights this problem and suggests a framework in which new studies should be conducted, but many of the underlying problems remain to be addressed. Perfusion data processing is based on analysis of the signal-time curve produced as the contrast bolus passes through the cerebral circulation (Fig. 6). This curve can be processed in many different ways to extract parameters that reflect either the CBF, cerebral blood volume (CBV) or mean transit time (MTT). These three parameters are linked by the equation CBV = CBF × MTT, otherwise known as the ‘central volume principle’. This is equivalent to saying that the number of people in a building at any one time (the volume) is equal to the number of people entering minus the number of people leaving the building (the flow) per unit of time in which this is measured.

The simplest way of displaying the perfusion information is to provide images of relative perfusion values, so that any stroke lesion can be seen as an area of reduced flow relative to the normal hemisphere. With more complex mathematical processing [30], quantitative perfusion values close to absolute values found in PET imaging [70, 71] and experimental models [1] can be produced. Unfortunately several assumptions need to be made in processing the data, the calculation of flow is influenced by the ‘tightness’ of the bolus which is affected by the rate of injection and poor cardiac output [72], by how the start and end-points of the curve are determined, the arrival of contrast may be delayed by factors such as carotid stenosis, and areas of reduced flow may result in signal levels that are below the noise level in the scan [73], all of which increase variability and limit the reliability of perfusion imaging [74]. In addition, it is likely that flow levels are in reality highly heterogeneous.

Whilst considerable progress has been made in understanding how to reduce the impact of these variables on the resulting perfusion values, we are still a long way from having a standardized technique that can be used reliably to identify penumbra in stroke. For example, CBF and CBV lesions are typically smaller than MTT lesions regardless of which parameter is used to represent each of these [43, 75] and of whether relative or fully quantitative parameters are used [76]. There is no apparent advantage for fully quantitative parameters compared with relative indices [70, 76, 77], and relative indices are more readily available and less computationally intensive. Processing the same data set to give images of 10 different perfusion parameters produces 10 different-sized perfusion lesions [76], even though these parameters can be grouped into those reflecting CBF, CBV and MTT, so a maximum of three lesions should be produced [76, 78]. What is particularly worrying is that the different-sized perfusion lesions result in very different proportions of patients appearing to have diffusion-perfusion mismatch (i.e. salvageable tissue), ranging from 10% to 70% depending on which perfusion parameter was used [76]. Without standardization of the mismatch definition and measurement across all stroke centres, this would mean that there might be huge variation in the proportion of patients offered treatment or even considered eligible for some trials in different centres.

Many studies have sought to determine perfusion thresholds for dead/salvageable versus not-at-risk tissue, with MR, CT and other techniques [67]. No consistent threshold was identified. In an update of this systematic review focusing on MR and CT perfusion imaging, we first found widely differing definitions of lesion core (e.g. these included an ADC threshold-delineated lesion, an outline of the visible DWI lesion, a final infarct on T2, tissue in both the acute DWI and the final T2 or based on PET criteria) and of the penumbra (e.g. these included DWI expansion, final T2 lesion minus acute DWI, mismatch that infarcted, mismatch that survived, acute DWI that did not infarct or PET criteria). It is unfortunate that such important tissue categories are so inconsistently defined. As for actual thresholds, despite starting with over 100 papers, very few studies provided information that was likely to be reliable. For example, reliable data on the penumbra-to-core threshold on MR come from two studies, one of nine and the other of 14 patients. For CT perfusion, six papers provided data on the not-at-risk-to-penumbra boundary from about 100 patients but this included a mixture of patients who did or did not receive thrombolysis, and in whom reperfusion was and was not accounted for, making it virtually impossible to interpret these threshold data in any meaningful way. Therefore, the second key finding from the systematic literature review was widely differing perfusion values in each of these tissues [67]. Although the median thresholds are similar to those identified in experimental studies, the standard deviations are wide, meaning that perfusion imaging thresholds are too variable for clinical decision-making. This variability is summarized in Fig. 7. It is not just a problem of choosing which perfusion parameter to use, because despite apparently performing the same analysis, passing the same data set through different manufacturer’s software and bespoke academic software can also produce different sized perfusion lesions for the same parameter (Dani K, Thomas R, Chappell F, Muir K, Wardlaw JM, unpublished data). There is clearly an urgent need for standardization of perfusion imaging.

Figure 7.  Critical blood flow levels derived from experimental and human studies for cell death, survival in a shut down but viable state, and not at risk tissue, with an indication of how diffusion and perfusion image data correspond to these critical levels. Although many studies agree on critical levels of perfusion, the large variation in standard deviation of perfusion values means that there is large variation in perfusion values across the ischaemic lesion meaning that discrete thresholds are nonspecific [67, 71].

Finally, there is little information on observer reliability of either visual or computerized diffusion lesion volume assessment [56, 79, 80], or of the assessment of perfusion lesions in stroke [81, 82]. Visual assessment of diffusion [83] or perfusion [18, 84] lesion extent using a scoring system like ASPECTS is probably as reliable as the more tedious lesion volume measurement. The limited information available suggests that the observer variability for lesion volume measurement is between 10% and 25%, i.e. substantial. A 20% variation in observer reliability either on visual or computer assessment could easily result in some of the apparent variation in results between different studies. On axial images, it is quite difficult to judge accurately by eye if a perfusion lesion is 20% or more larger than a diffusion lesion by volume.

Why has not penumbral imaging worked so far?

It was hypothesized that MR imaging of DWI/PWI mismatch would show penumbra (or ‘tissue at risk’) over 10 years ago. Some stroke physicians use the presence of mismatch between three and six or more hours after stroke as an indication for nonlicensed use of thrombolysis, but this approach has yet to be fully validated for all the reasons given above [16–20, 85]. There is large variation in perfusion threshold values, and ADC thresholds do not reliably identify core tissue.

What, in addition to inconsistent definitions and processing techniques, might explain this variation? Have we failed to recognize the true heterogeneity of the acute ischaemic lesion (Fig. 7)? Any analysis of diffusion or perfusion data based on thresholds, and without careful operator scrutiny and editing (Fig. 4), will include large amounts of normal tissue in addition to the abnormal tissue – in receiver operator characteristic analyses of perfusion data, these many normal values cause bias [71]. There are also several physiological reasons which have largely gone unrecognized.

First, normal blood flow and ADC values [86] differ between grey than in white matter, but few analyses segment the abnormal tissue into grey or white matter to assess their ADC or flow values independently. Indeed, in abnormal tissue this would be very difficult to do automatically because the ischaemic tissue signal changes effectively remove the segmentation boundaries. However, averaging flow values across a lesion means that a lesion predominantly in white matter will have lower flow than a lesion predominantly in grey matter just by virtue of its anatomy, because grey matter flow is inherently higher. All experimental animals have a substantially lower white : grey matter ratio than humans (about 10 : 90 in rats versus 45 : 55 in humans), meaning that the tissue thresholds derived from experimental models really only apply to grey matter.

Secondly, there is substantial individual variation in the distribution and size of the borderzones between the main arterial territories (in some patients there is even side to side variation) and adequacy of collateral arterial supply (Fig. 8). Thus, in any one individual, depending on the adequacy of collateral pathways between the major intracranial arteries and therefore the size of their borderzones, it will be difficult to tell whether their infarct occupies the whole of their MCA territory or just a small part of it, and therefore whether for that individual the amount of mismatch is large or small [87, 88].

Figure 8.  An example to illustrate how variation in arterial borderzones affects apparent penumbral tissue. In the right hemisphere, there are good collateral pathways between the middle cerebral artery (MCA), anterior cerebral artery (ACA) and posterior cerebral artery (PCA) resulting in large borderzones (blue shading) between these territories where blood supply could be maintained by the ACA or PCA if the MCA became occluded. In the left hemisphere, the collateral pathways are poor with small borderzones (yellow shading), meaning that the territory which is completely reliant on the MCA is much larger, twice the size by volume. Thus a proximal MCA occlusion in the right hemisphere would tend to produce a smaller amount of core infarct (represented by the white area) and might appear to have a smaller amount of penumbra at a given time after stroke than a proximal MCA occlusion in the left hemisphere. On the right, the amount of at-risk tissue might look quite small if flow in the borderzone was well maintained, resulting in the impression of little at-risk tissue. In the left hemisphere, the poorer collateral pathways and smaller borderzones might result in a much larger area of apparently at-risk tissue as represented by the perfusion defect. However what we actually see would be very dependent on the site of MCA occlusion versus the extent of the borderzone and whether there has already been some reperfusion. The borderzones vary substantially between individuals and even between hemispheres in one individual [88, 114], and it is not possible to tell how large any one person’s borderzones actually are in life. Thus it is impossible to judge whether any one individual stroke patient has a large or small amount of truly penumbral tissue based on mismatch imaging.

Thirdly, work in cats has shown that waves of spreading depolarization (a well-recognized feature of ischaemic brain) are associated with waves of vasospasm spreading across the lesion [89]. Each wave of vasospasm results in a transient and profound drop in blood flow which recovers although never quite returns to its prevasospasm level. Over time, repeated waves of vasospasm result in steadily declining blood flow, further compromising the tissue [90]. These may have lower periodicity in the human brain than in the cat, but there is no reason to think that they do not also occur in humans. A perfusion scan performed during or just after a spreading depolarization-induced wave of vasospasm would result in apparently lower perfusion than would have been seen 10 min earlier or later, thus emphasizing the transience and constant variability of brain haemodynamics, a factor which is not appreciated from individual ‘snapshot’ perfusion images.

However, there is a further problem with mismatch. Few studies have included patients without mismatch, but those that did suggested that up to about 50% of patients without DWI/PWI mismatch have some degree of infarct growth [75, 91, 92]. However, there are too few patients without mismatch in these studies to produce reliable data on whether the amount of infarct growth in patients without mismatch is as much as in patients with mismatch, but it is likely infarct growth is largest in patients with mismatch. Even so, even a small amount of infarct growth in patients without mismatch would suggest that there is some scope for treatment to reduce the final amount of brain tissue damage in all patients, without as well as with mismatch.

There is insufficient room in this review to deal with a more detailed critique of perfusion imaging and mismatch. Suffice it to say that, for all the methodological and physiological reasons given above, the mismatch approach has limitations [18, 20, 85, 93]. The most practical use of perfusion imaging, especially with CT perfusion, may simply be as a way of increasing confidence in the diagnosis of ischaemic stroke – if early ischaemic signs are difficult to see on CT especially to the less experienced observer, then seeing a large perfusion defect may help establish the diagnosis of ischaemic stroke [93]. Similarly, seeing an occluded artery on CT angiography (CTA) or MR angiography (MRA) would increase the confidence in the diagnosis of ischaemic stroke. The converse however is not true – not seeing an occluded artery or a perfusion lesion does not mean that the patient is (i) not having an ischaemic stroke or (ii) would not benefit from treatment.

Other factors associated with infarct expansion

If mismatch is not the only factor that is associated with infarct growth, what other factors might be? All this suggests that the patient’s response to ischaemia, not just the initial insult, may also be important in determining outcome. Indeed several patient-specific factors are emerging. Infarct growth is associated with older age [94, 95] and leukoaraiosis [96, 97] (a feature of older age). The reasons for the association with age are not entirely clear, unless they are because of an association with less ability to withstand reduced perfusion, or to autoregulate, in older patients. Leukoaraiosis is associated with reduced CBF, perhaps meaning that older, leukoaraiotic brains are less able to withstand fluctuations in blood flow, less able to vasodilate, have poorer collateral pathways or are more prone to hypoxia for other reasons [98].

Alternatively, perhaps older patients are less able to spontaneously lyse the occluding parent artery thrombus. Or they may be less able to prevent microvascular thrombus propagation extending from the parent artery occlusion into the smaller vessels involved in the infarct. For example, elevated plasma levels of the inflammatory marker C-reactive protein (CRP) are associated with infarct growth [99, 100]. Elevated levels of fibrinolysis inhibitors on admission are associated with delayed thrombus lysis and slow reperfusion [101]. Plasma markers of thrombosis activation (e.g. fibrin D-dimer) and inflammation (e.g. CRP and interleukin-6) are associated with poor outcome after stroke and with increased risk of death from stroke and all causes [102], whether or not patients receive thrombolysis [103].

These markers may partly indicate tissue injury, but probably also indicate a generally poor vascular state. The mechanism of poor outcome and increased death not just from stroke but from all causes [102] is not well understood, but they might contribute to worsening of the stroke through delay in thrombus lysis, promotion of large artery re-thrombosis or microvascular occlusion [104] thereby preventing reperfusion of tissue even when the parent artery recanalizes, the long overlooked ‘no re-flow phenomenon’ (Fig. 9).

Figure 9.  The ‘no reflow phenomenon’. (a) The artery is occluded by thrombus, some very limited flow may persist in the arterioles but the tissue becomes ischaemic. (b) As astrocytes and other perivascular cells start to swell absorbing fluid from the plasma, the arterioles narrow and the blood within them also clots. (c) The perivascular cells swell more, pressing on adjacent arterioles and capillaries and restricting flow in adjacent tissues and increasing the extent of ischaemic damage. Now, even if the parent artery thrombus dissolves, tissue perfusion is unlikely to be restored at the arteriolar and capillary level because the vessels are so narrowed by swollen perivascular tissues.

The ‘no re-flow phenomenon’ was originally described in experimental models in rabbits in 1967 [105–107], but dropped out of sight as an important mechanism in infarct evolution until recent observations [95] amongst other factors have led to a resurgence in recognition of this important phenomenon [108, 109]. Perhaps the intense focus on assessing patients within the first few hours has limited awareness of important processes that occur later, that may worsen outcome and which could offer further opportunities for intervention.

The ‘no-reflow phenomenon’ was originally used to describe the observation that the ischaemic process can cause vascular changes severe enough to impair the return of blood flow to the tissue even if the parent artery reopens [105]. Thus, even a short period of ischaemia insufficient to kill the brain cells could prove indirectly lethal to these cells. In the original experiments in rabbits, it was noticed that total arterial occlusion for up to 15 min, followed by reperfusion at arterial pressure with carbon black, resulted in failure of the carbon black to enter all parts of the affected arterial territory. The failed (white) areas became larger with increasing duration of ischaemia. Histological examination showed the arterioles and capillaries in these areas to be filled with red cells. This was not inhibited by preocclusion flushing with heparin and no platelet-fibrin thrombi were observed. Instead, electron microscopy demonstrated swelling of the perivascular astrocyte cells, their greatly hydrated perivascular feet bulging into the capillaries and reducing the lumen to a narrow slit. As the perivascular astrocytic end feet swell, they draw fluid from the interstitial compartment as well as from plasma. This would appear on diffusion imaging as the earliest hyperintense diffusion signal (ADC fall) and would correspond with the glial cell swelling seen histologically in animal studies [65]. The fluid absorption into the astrocytes leads to further concentration of the macromolecules in the blood and increased viscosity, further impeding the chances of successful tissue reperfusion when the parent artery recanalizes (Fig. 9) [105, 106].

What is the patient equivalent of the ‘no-reflow phenomenon’? We noticed that delayed normalization of the diffusion signal (persistently hyperintense signal) in the subacute phase occurred in 50% of patients with large MCA infarcts (e.g. Fig. 5), and was associated with impaired areas of tissue reperfusion within the infarct and worse clinical outcome than in patients with early normalization of the diffusion signal, despite both groups having similar blood flow levels acutely [95]. This delayed normalization of the diffusion signal, for weeks in some cases, in the presence of persistently reduced blood flow, suggests ongoing cell oedema impeding tissue reperfusion. This in turn could impair blood flow to areas at the periphery of the lesion that might otherwise have survived, could increase inflammation and lead to lesion growth as adjacent areas that were not affected by the initial ischaemia are progressively recruited into the lesion. This could explain infarct growth in patients without mismatch at presentation. The association with worse clinical outcome would be consistent with failed reperfusion which would otherwise have led to improved outcome. Waves of spreading depolarization with accompanying vasospasm would compound astrocytic swelling and would worsen capillary narrowing and ‘no-reflow’, explaining the observed progressive fall in tissue perfusion [89] and create a vicious cycle of increased swelling, falling flow, worse ischaemia, infarct expansion and worse outcome. The balance of perfusion versus swelling could explain why neurones continue to die for weeks after the ischaemic stroke [110]. Others have observed failure of reperfusion at the tissue level when the occluded artery has recanalized [108], and commented on the importance of distinguishing between arterial recanalization and tissue reperfusion to avoid further confusion over these different processes [109], but the importance of these subacute changes and the potential that they present for new therapeutic intervention have largely been overlooked and deserve greater attention.

What does this mean for stroke treatment?

First, it further underlines the fact that ischaemic stroke is a dynamic process that evolves over many hours, even days, offering many opportunities to intervene to break the vicious downward spiral of progressing tissue damage. There is increasing evidence of persisting viable neurones at up to 24, 48 [26] and even 52 h after stroke [81], and further evidence of ongoing tissue damage with neuronal death after that [110]. We should not write off all hope of acute intervention at 3, 4.5 or 6 h when in fact the damage is just beginning. Yes, restoring tissue perfusion as quickly as possible is the number one goal, but opening the parent artery occlusion is just the beginning.

From all that has been said thus far, it must be patently obvious that the damage is not all done in the first few hours, but that the acute stroke lesion is a dynamic process that goes on evolving over hours, days and possibly weeks after the stroke. The totality of the randomized thrombolysis trials shows potential benefit in selected patients up to 24 h [22] as well as the observational studies [21]. Different processes likely act at different times, but the net effect is a series of events whose risk is increased or decreased by immediately preceding events as well as by the patient’s underlying health state and key factors about the stroke. Time from symptom onset alone is almost certainly too crude for treatment decisions, and different decisions are probably required at different stages, probably guided by information from imaging. For some patients, 3 h may already be too late, whereas for others there may be important amounts of penumbral tissue for 9 h, possibly even 12 h after stroke that would benefit from recanalization therapies.

Thereafter, whilst the debate on the merits of penumbral imaging continues [20, 85, 111, 112], efforts should be focused on finding ways of diminishing the likelihood of no reflow, maintaining vascular patency at the capillary level, reducing lesion swelling and preserving adjacent tissue. Factors that contribute to failure of tissue reperfusion at later times present a target for therapeutic intervention which could start hours after the acute stroke, widening the therapeutic window to those patients who still do not reach hospital within the first few hours. Agents which prevented arteriolar re-thrombosis after successful recanalization, or can reduce microvascular occlusion, even simple factors like adequate hydration, could provide important additive benefits after recanalization therapies, or be used in patients who arrived too late for thrombolysis to be used safely. It is long overdue that we start thinking of acute stroke treatment as a continuum, that we recognize the dynamic nature of the brain lesion and adopt a more holistic approach to understanding the pathophysiology, rather than just giving up after the first few hours.

Conflict of interest statement

J.M.W.’s department received funding from Boehringer Ingelheim for expert scan interpretation for the ECASS 3 Trial.

Acknowledgements

J.M.W. is part funded by the Scottish Funding Council through the SINAPSE Initiative (Scottish Imaging Network, A Platform for Scientific Excellence, http://www.sinapse.ac.uk).