Magnetic resonance imaging (MRI) is considered an integral part of the diagnostic evaluation performed in patients presenting with symptoms suggestive of stroke. MRI performed in acute stroke setting is not only superior to computed tomography in detection of ischemic lesions, but can also accurately identify acute or chronic hemorrhage (Chalela et al. 2007). Other advantages of MRI over computed tomography include a better discriminative performance in differentiating cerebrovascular events from other causes leading to focal, sudden-onset neurologic deficits (Fiebach et al. 2002) and identification of acute ischemic lesions at very early time points with better sensitivity (Mullins et al. 2002). All of these advantages, together with increasing availability of the technology, make MRI the ideal diagnostic tool in the setting of acute stroke. However, the utility of MRI for the clinician is not only limited to its diagnostic role. Ongoing technological advances in MRI technology and advent of new imaging techniques have now made it possible to use MRI as a prognostic tool in ischemic stroke.
MRI as a predictor of tissue outcome
The most classical information obtained from MRI while making projections regarding final tissue outcome comes from extrapolation of the extent of initial ischemic insult to final lesion volume. Conventional MRI sequences like T1, T2, and fluid-attenuated inversion recovery (FLAIR) become sensitive to ischemic changes only after a net increase in water content of the cerebral tissue and therefore can detect ischemia after a few hours of symptom onset. Diffusion-weighted imaging (DWI), on the other hand, is sensitive to cytotoxic edema—the primary pathology during the hyperacute ischemia setting—and can therefore provide the opportunity to determine the extent of ischemic injury even within the initial hours of ischemia. Volumetric studies have shown a perfect correlation ranging from 0.84 to 0.99 between acute ischemic lesion volume on DWI and final lesion volume (Lovblad et al. 1997; Tong et al. 1998; Beaulieu et al. 1999; Schaefer et al. 2002). However, a perfect correlation does not translate into accurate identification of final lesion volume by only using DWI data on a case-by-case basis. Studies evaluating lesion volume dynamics by serial MRI examinations highlight that DWI lesion volume underestimates final infarct size in majority of patients and there is a growth by 144%–180%, on average, in the size of ischemic lesions on follow-up (Warach et al. 2000; Kidwell et al. 2004; Kranz and Eastwood 2009). On the other hand, in approximately a quarter of patients, there is evidence for some degree of DWI reversal (Kranz and Eastwood 2009). All of these findings suggest the presence of factors, other than extent of cytotoxic edema, in determination of tissue fate after an ischemic insult.
One of these major factors is the amount of hypoperfusion within the ischemic territory, which can be assessed by perfusion-weighted imaging (PWI) MRI. Albeit PWI cannot assess the actual amount of cerebral blood flow (CBF) and determines tissue perfusion in relative terms, it still provides clinically reliable perfusion measures that are correlated with the gold-standard positron emission tomography (Takasawa et al. 2008). Brain tissue that appear normal on DWI, but have abnormal perfusion, are considered to represent regions that are viable, but at risk for conversion to infarction over the ensuing hours (Kidwell et al. 2004). This concept, known as the diffusion/perfusion mismatch model, can be considered as the MRI surrogate of ischemic penumbra, and provides the clinician with prognostic information regarding the risk for infarct progression (Fig. 1). The validity of diffusion/perfusion concept were supported by observations, which showed that patients with diffusion–perfusion mismatch were more prone to infarct growth, and establishing recanalization or reperfusion could be helpful in preventing infarct progression (Jansen et al. 1999; Parsons et al. 2002; Davis et al. 2008). Furthermore, volumetric analyses have demonstrated a significant correlation between final infarct size and various measures of perfusion, with cerebral blood volume showing the highest correlation followed by CBF and mean transit time (Schaefer et al. 2002). Nonetheless, there were discrepancies between CBF or mean transit time volumes and final infarct size, which primarily arose from overestimation of lesion size by these perfusion maps, compared to cerebral blood volume maps or DWI. This was considered a reflection of the heterogeneous nature of the mismatch tissue, with some regions at high risk for conversion to infarction, whereas other regions representing benign oligemia and practically very low risk for undergoing permanent injury. One approach to overcome such a limitation has been the use of more strict thresholds on perfusion maps that accurately represent tissue at risk; currently, a time to peak (or Tmax) threshold between 4 and 6 s is considered as the most optimal measure of critical hypoperfusion on PWI (Olivot et al. 2009). However, Tmax thresholds vary significantly depending on various factors like physiologic variables or image-acquisition parameters and relying on a single threshold during predictions, despite its practicality, might introduce a certain level of inaccuracy, especially in the real-world setting (Calamante et al. 2010). This shortcoming can partially be overcome by using probabilistic maps that take into the severity of Tmax abnormality across its whole range, rather than dichotomizing the perfusion abnormality at a single threshold (Nagakane et al. 2012). Still, volumetric assessments depending on a single perfusion parameter do not entirely represent the complex pathophysiology within the ischemic tissue. One method that was introduced to reflect the heterogeneity within the ischemic tissue has been the use of voxel-by-voxel prediction models (Wu et al. 2001). These models create probability maps for infarction on a voxel-by-voxel basis and can predict risk of infarction with high sensitivity and specificity, especially when both DWI and PWI data are taken into account (Wu et al. 2001, 2006). However, regardless of the prediction algorithm used, models that only depend on DWI and PWI can not entirely explain the variability observed in infarct progression.
Figure 1. The presence of a small diffusion-weighted imaging lesion (a) together with a large perfusion-weighted imaging lesion (b; MTT maps) is considered to represent the presence of salvageable ischemic tissue. Other important markers of tissue hypoperfusion are hyperintense vessels on fluid-attenuated inversion recovery (c; arrowheads) and hypointense vessels on gradient-echo recalled or susceptibility-weighted imaging (d; arrowheads).
Download figure to PowerPoint
The predictions for tissue outcome can be improved by incorporation of vessel status into the algorithms. Patients with large vessel occlusion evident on MRI angiography do not only have an increased probability of diffusion perfusion mismatch, but also are more prone to lesion expansion (Barber et al. 1999). More importantly, predictions for tissue outcome are greatly affected by the degree of recanalization; prognostic models developed in natural history models overestimate final infarct size in patients with successful recanalization (Rosso et al. 2009). Further information regarding vascular status of ischemic territory can be obtained by assessing the degree of collateral circulation on MRI. Hyperintense vessels observed on FLAIR images that are considered to reflect stagnant and slow flow of the antegrade or retrograde collaterals and hypointense vessels observed on gradient-echo recalled (GRE) or susceptibility-weighted imaging that are considered to reflect deoxygenated blood within the draining veins or collaterals represent the presence of significant perfusion deficit and thereby probability of infarct expansion (Fig. 1) (Hermier et al. 2005; Kaya et al. 2009; Lee et al. 2009).
Spatial information also plays an important role in the fate of ischemic tissue. Patients with insular lesions are more prone to infarct progression, when compared with patients with lesions sparing the insula (Ay et al. 2008a). In addition, regardless of the perfusion parameters, brain tissue that is in close proximity to DWI lesion has a higher risk to convert to infarction, whereas hypoperfused tissue located in the periphery of mismatch region is generally spared on follow-up images (Ay et al. 2006). Patterns of lesion progression within the penumbra region also differ with respect to territory involved in the cerebral parenchyma (Wu et al. 2009). Apart from these factors that are imminently related to the acute ischemic insult, more chronic insults to the brain parenchyma introduce variability to infarct progression within the mismatch region. One of these factors detectable by MRI, the leukoaraiosis burden, which reflects a composite measure of baseline brain integrity is closely associated with conversion of ischemic tissue to infarction independent of the level of hypoperfusion (Ay et al. 2008b).
Cerebral ischemia triggers a complex set of pathophysiologic events. Clinical MRI, which generally encompasses T1, T2, FLAIR, MRI angiography, DWI, PWI, and GRE helps us understand this complex process up to a certain extent. The development of new MRI sequences and image-processing algorithms might help us in explaining the variability in ischemic lesion growth, especially during the acute setting. pH-weighted imaging, which can be used to assess the degree of acidosis within the ischemic tissue (Sun et al. 2007), blood oxygen level-dependent MRI, which can be used as an indicator of oxygen extraction fraction (Geisler et al. 2006), iron oxide particle-enhanced MRI, which evaluates the inflammatory reaction in the ischemic territory (Saleh et al. 2007) are promising tools that might help us in the near future to tease out the heterogeneity observed in patients with acute ischemic stroke.
MRI as a predictor of clinical outcome
Clinical outcome after ischemic stroke depends on a number of factors, including the degree of the initial ischemic insult, the extent of recovery process and the presence or absence of recurrent events, among many others. MRI can provide important information regarding almost all of these main determinants of clinical outcome.
Lesion volume, whether assessed during initial hours of symptom onset by DWI or assessed at later time points by T2 or FLAIR is an independent and major predictor of clinical outcome after ischemic stroke (Thijs et al. 2000; Derex et al. 2004; Engelter et al. 2003; van Everdingen et al. 1998; Sanak et al. 2006). However, the correlations are far from perfect and are moderate at best, irrespective of the timing of MRI (Thijs et al. 2000; Derex et al. 2004; Engelter et al. 2003; van Everdingen et al. 1998; Sanak et al. 2006). Therefore, all the algorithms described in the previous section that are used in prediction of final infarct volume are also limited in this aspect in accurately providing clinical prognostic information on a case-by-case basis. One of the major factors contributing to this discordance is lesion location. Lesions located in clinically eloquent areas can have devastating results irrespective of their size, whereas large lesions might have minimal clinical implications if located in relatively silent locations. Therefore, new predictive models are being developed that make estimations by combining lesion volume and location information (Menezes et al. 2007). These models have significantly better predictive accuracies when compared to models that just rely on volume or location information. Predictions regarding contribution of lesion location to clinical outcome can further be improved by use of diffusion tensor imaging. For example, the assessment of white matter tract integrity within the corticospinal tract or dorsal/ventral language pathways by diffusion tensor imaging, either in the acute or chronic stage, can help refining predictions regarding the degree of motor and language dysfunction (Breier et al. 2008; Hosomi et al. 2009; Calamante et al. 2010; Lindenberg et al. 2010). Similarly, the activation patterns on functional MRI, an invaluable tool for studying stroke recovery, can also provide prognostic information regarding functional outcome, even when obtained during the initial days after stroke (Marshall et al. 2009). Various MRI approaches, like steady-state contrast-enhanced MRI, are also very promising tools under development that can be used to evaluate another aspect of recovery, angiogenesis (Seevinck et al. 2010).
Apart from the degree and location of the ischemic insult, prior lesion burden and inherent reserve of the brain to recover are important determinants of clinical outcome. The magnitude of chronic infarcts and leukoaraiosis as detected by T2-weighted or FLAIR images are independent predictors of unfavorable clinical outcome after ischemic stroke (Arsava et al. 2009; Putaala et al. 2011). Although it is currently unknown by which specific mechanisms these imaging markers affect outcome, they seem to reflect a composite measure of vascular, neuronal, and glial integrity within the cerebral tissue, and thereby play a role at numerous stages of ischemia ranging from infarct evolution during initial hours to plasticity over the following months of stroke (Ishihara et al. 1999; Ay et al. 2008b; Grefkes et al. 2008).
Recurrent ischemic episodes after the index event introduce significant disability and thereby are associated with unfavorable clinical outcomes (Sacco et al. 1989; Moroney et al. 1998). MRI is an important tool in estimating the recurrence risk after both ischemic stroke and transient ischemic attacks (TIA). The risk of recurrent stroke is highest during the initial 3 months (Burn et al. 1994; Sacco et al. 1994; Petty et al. 1998) and the clinical tools that are used to predict recurrence risk are only validated for assessing the long-term, rather than short-term risk (Kernan et al. 2000; Weimar et al. 2009). Certain lesion characteristics on MRI, however, can guide the clinician in identifying patients more likely to harbor a recurrent event in the early period after stroke. Multiple acute infarcts on DWI, acute or subacute infarcts simultaneously present in different circulations, presence of infarcts of different ages and isolated cortical infarcts are considered as tissue signatures of an unstable stroke etiology and thereby indicate patients at high risk for recurrence (Lansberg et al. 2001; Wen et al. 2004; Bang et al. 2005; Sylaja et al. 2007). A prognostic tool that combines these features together with clinical variables is now available and can be used in prediction of early risk of recurrence both in patients with ischemic stroke and DWI-positive transient neurologic symptoms (Ay et al. 2010; Arsava et al. 2011). Apart from the lesion characteristics on MRI, the presence or absence of acute ischemic lesions on DWI provides important prognostic information in patients with TIA; the risk of recurrent stroke is almost 20-fold higher in DWI-positive compared to DWI-negative patients (Giles et al. 2011). Combining this imaging information with validated clinical tools used to predict early stroke recurrence in TIA patients dramatically improves the predictive accuracy of the models (Giles et al. 2010; Merwick et al. 2010). Because of this highly distinctive risk profile among DWI-positive and -negative patients, American Heart Association now proposes to use a tissue-based definition for TIA, and define patients with transient neurologic symptoms as TIA only if DWI is negative and as stroke if DWI is positive for acute ischemic lesions (Easton et al. 2009).
The role of MRI in therapeutic decision algorithms
The ultimate goal of all these tissue or clinical prediction algorithms is to tailor therapies on an individual basis and optimize the risk–benefit ratio of therapeutic approaches used in the acute and chronic stages of ischemic stroke.
The current therapeutic approach in acute ischemic stroke relies on successful recanalization of the occluded artery to establish reperfusion within the ischemic territory. Therefore, it is highly critical to identify patients that are more likely to benefit from recanalization/reperfusion therapies. The ideal patients in this regard are those with a small infarct core, large salvageable penumbra, and low risk for intracerebral hemorrhage. The presence of a small lesion on DWI, together with a large DWI/PWI mismatch might therefore be considered as an imaging surrogate of this ideal clinical scenario where thrombolytic therapy could be beneficial. This hypothesis was tested in the observational DEFUSE trial where patients underwent MRI prior to receiving intravenous tissue plasminogen activator treatment 3–6 h after symptom onset (Albers et al. 2006). This study revealed the presence of two different imaging profiles; (i) patients with an initial DWI lesion volume of ≥ 100 mL and/or a PWI lesion volume of ≥ 100 mL (defined as ≥ 8 s of Tmax delay) comprised the malignant profile where reperfusion did not provide any clinical benefit and more importantly was associated with an increased risk of intracranial hemorrhage, and (ii) patients with an initial DWI lesion volume of 10–100 mL together with a PWI lesion volume of ≥ 10 mL and 120% or more of the DWI lesion comprised the target mismatch profile and represented the group of patients where early reperfusion was beneficial. Other small-sized observational studies have tried to come up with lesion volume thresholds to identify patients that might benefit from intraarterial thrombolysis and suggested an initial DWI lesion volume ≥ 70 mL as a marker of unfavorable prognosis (Yoo et al. 2009). Two separate randomized trials, DIAS-2 and EPITHET, assessed the utility of DWI/PWI MRI in patient selection for thrombolytic therapies, but unfortunately were not able to show a statistically significant benefit of the active treatment in patients with diffusion–perfusion mismatch (Davis et al. 2008; Hacke et al. 2009). There were, however, some issues with definition of mismatch region in both trials, and regardless of the treatment received, the beneficial effect of reperfusion on tissue and clinical outcome was evident in EPITHET. Another important concern in thrombolytic treatment is the risk for hemorrhagic conversion. Permeability MRI, by assessing the integrity of blood–brain barrier, is a promising tool in identifying patients that are at risk for this devastating complication (Bang et al. 2007). Ongoing trials will shed more light on the role of DWI/PWI in patient selection for intravenous or intraarterial thrombolytic therapies.
The therapeutic guidance of MRI is not only limited to patient selection for thrombolysis. In the acute setting, the identification of large lesions, specifically a DWI lesion volume ≥ 145 mL is a highly sensitive predictor of malignant middle cerebral artery infarction and therefore can be considered as a selection criteria for early hemicraniectomy (Oppenheim et al. 2000). Leukoaraiosis burden on FLAIR or T2, and cerebral microbleed burden on GRE or susceptibility-weighted imaging are predictors of intracranial hemorrhage in the long term and can be used to tailor the intensity of anti-platelet/anti-coagulant therapy following ischemic stroke (Smith et al. 2002; Fan et al. 2003). More recent developments like high-resolution plaque imaging or fibrin imaging might prove to be useful in the near future in identification of high-risk atherosclerotic plaques (Spuentrup et al. 2008; Oppenheim et al. 2009).