Introduction

Stroke is one of the primary causes of adult mortality, morbidity, and disability [1, 2]. Ischemic stroke is caused by a vascular blockage that results in sudden and severe hypoperfusion, leading to neurologic dysfunctions and brain tissue injury. The most severely hypoperfused brain tissue forms a core of irreversible damage (infarction), and the surrounding hypoperfused area is at risk of infarction (penumbra). If cerebral perfusion is not restored promptly, the infarction core may expand over time to the hypoperfused territory [3]. The goal of acute stroke treatment is to rapidly and safely recanalize the occluded vessel(s), salvage ischemic tissue, prevent infarct growth, and minimize hemorrhagic complications [4,5,6].

The Transition from Stroke Onset Time Clock– to Tissue Clock–Based Treatment

Intravenous tissue plasminogen activator (IV tPA) administered within the first 4.5 h of stroke onset has been and continues to be the standard care of acute stroke treatment [7]. tPA restores perfusion and improves functional outcomes by inducing fibrin degradation to break down the blood clot. However, tPA’s therapeutic time window is very narrow, limiting it to a very small group of stroke patients. Although time is critical, there has been tremendous interest in individualizing stroke therapy by transitioning from stroke onset time–based treatment (time clock) to salvageable tissue–based patient enrollment (tissue clock) [8, 9]. The key is to identify stroke patients with substantial penumbra tissue who are likely to benefit from reperfusion treatment [10]. Recent DAWN [11] and DEFUSE-3 [12] trials have shown that imaging-guided late-window recanalization is beneficial in carefully selected large vessel occlusion (LVO) stroke patients. Although routine stroke imaging has been well established in stroke clinical trials, it has been recognized that there are limitations. Patients in the DAWN trial had relatively small infarctions (average core < 10 ml). The majority of stroke patients (~ 70%) presenting 6–24 h with NIH stroke scale (NIHSS) over 6 are not DAWN and DEFUSE-3 eligible [13]. As Fisher and Xiong pointed out, it is urgent to determine both the lowest Alberta stroke program early CT score (ASPECTS) and the largest ischemic core volume where thrombectomy is no longer beneficial [14]. The ASPECTS is calculated by deducting 1 point from a score of 10 points for any evidence of early ischemic change for each of the defined regions, which provides a quantitative topographic CT score of acute stroke.

MRI has also been recently shown in the WAKE-UP trial to be useful in guiding tPA thrombolysis in stroke patients with unknown stroke onset time [15]. In this study, the presence of diffusion-weighted imaging (DWI) lesion and the absence of parenchymal hyperintensity on fluid-attenuated inversion recovery (FLAIR) MRI were used as a surrogate marker for early infarction amenable to thrombolytic therapy. However, many patients with early infarcts may be excluded by this approach since FLAIR hyperintense lesion can occur in 15% of patients imaged under 3 h and in 41% of patients imaged at 3 to 4.5 h [16]. A more accurate tissue marker has tremendous potential to improve selection in patients with unknown stroke onset time. Furthermore, a more tissue-specific imaging marker may also be helpful in evaluating strokes that do not arise from anterior circulation large vessel occlusions. For example, lacunar-type and posterior fossa infarctions may have different imaging characteristics than anterior circulation territorial infarctions [17].

Stroke Imaging

Imaging has played a crucial role in identifying acute stroke patients for thrombolytic and endovascular treatments [18]. There are four essential tasks of acute stroke imaging: to detect the presence of hemorrhage, to identify the location and severity of occlusion(s), to measure infarct core volume, and to estimate the penumbral tissue [10, 19]. Both CT and MRI have advantages and disadvantages in stroke imaging [20,21,22,23,24]. CT is currently the most commonly used imaging modality due to its wide availability and rapid acquisition. On the other hand, MRI is versatile and can characterize the hemodynamic, metabolic, and structural status of the ischemic tissue, which provides a comprehensive characterization of the ischemic core and penumbra [25, 26]. Whereas MRI exams may take longer than CT, it is worth noting that fast stroke MRI protocols with good diagnostic quality have been performed in scan times rivaling that of CT protocols for the evaluation of acute stroke patients [27].

Acute ischemia induces a cascade of tissue changes (Fig. 1), depending on the level and duration of hypoperfusion [28,29,30]. Ischemic tissue initially suffers from disrupted gene expression and protein synthesis under the condition of mild hypoperfusion (35–50 ml/100 g/min). Such changes are, however, not detectable using non-invasive imaging. With the further reduction in perfusion, ischemic tissue transitions from aerobic to anaerobic glycolysis and causes changes in cerebral metabolic rate of glucose (CMRG) and lactate production (25–35 ml/100 g/min). Although MR spectroscopy, such as 1H and phosphorous 31P MRS, can detect such metabolite changes, the spatiotemporal resolution of MRS is not sufficient for acute stroke imaging [31,32,33,34]. Upon metabolic disruption, ischemic tissue becomes acidic (lactic acidosis) with glutamate release, which represents a narrow range of perfusion thresholds between selective neuronal loss (25–50 ml/100 g/min) and infarction (under 22 ml/100 g/min) [35,36,37,38,39]. With a further reduction of perfusion level, key tissue metabolites such as phosphocreatine and adenosine triphosphate (ATP) deplete, which quickly leads to irreversible tissue injury and infarction. It is critical for stroke imaging to capture events along this cascade of worsening tissue damage so that the ischemic tissue can be staged correctly in real time for individualizing stroke therapy.

Fig. 1
figure 1

A diagram of ischemic tissue injury cascades following acute stroke. Note that the glucose metabolism, lactate, and pH changes occur before infarction, suggesting that such indices are biomarkers for defining ischemic penumbra (Markus HS. J Neurol Neurosurg Psychiatry. 2004;75(3):353–61)

Full size image

Diffusion and Perfusion MRI

DWI is currently the operational gold standard for defining the ischemic core [8, 40]. It has been documented that the ischemic core has a reduced apparent diffusion coefficient (ADC) for the first 7 days, followed by ADC normalization and subsequent increases above that of healthy tissue [41]. Within the first hours of stroke onset, DWI is the most sensitive and specific means of depicting the extent and size of infarction. It has been relied upon to establish the infarction volume thresholds in which recanalization will likely be futile and risky if above [42, 43]. Perfusion-weighted MRI (PWI) can be performed using either contrast-enhanced or non-contrast methods to depict the severity of ischemia [44,45,46,47,48]. Whereas arterial spin labeling (ASL) perfusion MRI provides superior perfusion measurement, it is often technically challenging in the acute stroke setting, and dynamic susceptibility contrast (DSC) MRI is more widely used [49,50,51]. In the absence of revascularization, the infarction may grow from the initial DWI lesion to approach the PWI lesion. It is worth noting that not all hypoperfusion lesions will proceed to infarction, likely because the collateral circulation can sustain the ischemic tissue and, therefore, slow down or even prevent infarction growth [52]. Altogether, DWI and PWI lesion mismatch has been postulated as an operational penumbra to identify stroke patients for reperfusion treatment in an extended time window [53].

Limitation of Routine Diffusion and Perfusion MRI

Although diffusion and perfusion imaging has been widely used, the perfusion/diffusion (PWI/DWI) lesion mismatch paradigm is oversimplified. The mismatch could not adequately depict the heterogeneity of the viable ischemic tissues [54]. The PWI/DWI mismatch may not only contain a mixture of penumbral and benign oligemic tissue but also fail to include a portion of the reversible diffusion lesion. Specifically, the perfusion thresholds for ischemic injury varies with sex, age, and tissue type (gray versus white matter) [55,56,57,58]. The perfusion lesion often overestimates the penumbra by including the mild ischemic area unlikely to infarct (benign oligemia) [59]. On the other hand, the DWI lesion suffers from graded metabolic disruption and could overestimate the ischemic core [60,61,62]. A portion of the DWI lesion is reversible with early thrombolysis, even in cases with large DWI lesions [63,64,65,66]. Whereas DWI reversal had been considered infrequent [67], recent studies revived the concept of DWI renormalization in patients with early recanalization [68]. For example, Hsia et al. pointed out, “Apparent diffusion coefficient (ADC) evolution in patients with early, complete revascularization, now more commonly seen with endovascular therapy, is strikingly different from our historical understanding.” [69]. The study concluded that recanalization and reperfusion lead to an earlier increase in intensity and a more rapid ADC normalization of the ischemic core than before. Early revascularization and ADC normalization often occur together, which may serve as a potential biomarker for developing future adjunctive treatment.

The Refined Mismatch Paradigm

Building on the initial concept of perfusion/diffusion lesion mismatch, Kidwell et al. proposed a modified mismatch paradigm to refine the imaging definition of the ischemic penumbra [52]. The refined penumbra has its outer boundary smaller than the perfusion lesion, so the hypoperfused benign oligemia is excluded from the penumbra. Also, the modified penumbra’s inner boundary extends into the diffusion lesion to include a portion of diffusion lesion that is potentially salvageable [70]. Although the modified mismatch paradigm based on clinical observation is highly plausible, there has been a lack of stroke imaging techniques to reliably characterize the heterogeneous ischemic tissue to refine the penumbra definition. The development of advanced stroke imaging techniques could improve the identification of salvageable tissues, leading to the development of new clinical protocols and monitoring therapeutic strategies.

Emerging Stroke Imaging Methods—Diffusion Kurtosis Imaging

The ADC calculation assumes a monoexponential decay of MRI signal versus diffusion b value [71,72,73]. However, diffusion in biological tissue does not precisely follow a Gaussian free diffusion profile due to displacement restriction and barriers [74]. Kurtosis is an index that describes the degree of non-Gaussian diffusion that has been overlooked in routine diffusion MRI. Diffusion kurtosis imaging (DKI) quantifies not only the diffusion rate (i.e., diffusivity) but also the degree of deviation from the Gaussian diffusion profile (i.e., kurtosis) [75,76,77,78]. In a study of acute/subacute ischemic stroke patients, Hui concluded that ischemia preferentially alters the intra-axonal environment and proposed focal enlargement of axons known as axonal swelling or beading as a potential mechanism for kurtosis change following stroke [79].

It has been shown that in a transient middle cerebral artery occlusion (MCAO) rat model, the DWI lesion without kurtosis abnormality renormalizes after early reperfusion. In contrast, the kurtosis lesion within the DWI lesion shows a poor response to reperfusion [80]. Briefly, Fig. 2 shows mean diffusivity (MD) and kurtosis (MK) images before and after reperfusion in rats, documenting partial diffusion lesion renormalization. The MD lesion was considerably larger than the MK lesion during MCAO. After reperfusion (90-min MCAO), the MD lesion partially reversed to about the same size as the acute MK lesion during MCAO. In contrast, the MK lesion had little change before and after reperfusion. This observation suggests that the routine DWI lesion is heterogeneous, and DKI may distinguish the irreversibly damaged infarction core from the portion of potentially reversible DWI lesions. Such experimental stroke finding is consistent with the clinical observation that early recanalization in acute stroke patients often results in partial DWI reversibility that is correlated with functional outcomes [64, 81]. Although the routine DKI protocol requires a minimum of 31 scans, a fast DKI protocol has been developed that requires 13 scans, reducing the scan time by over 50% [82,83,84]. The diffusion/kurtosis mismatch region has shown a trend of higher perfusion than the infraction core [85]. Because it has been reported that thresholding ADC does not predict DWI reversibility [86], DKI and DWI likely capture different aspects of ischemic tissue injury, complementing each other. The use of DKI to define infarction core potentially avoids the overestimation of irreversibly damaged infarction tissue and allows for an accurate depiction of ischemic penumbra for EVT in an extended recanalization window.

Fig. 2
figure 2

DKI predicts diffusion renormalization after early (90-min MCAO) reperfusion (Cheung et al. Stroke 2012:2252–4)

Full size image

Emerging Stroke Imaging Methods—Tissue pH

Acidosis is associated with oxygen–glucose deprivation and is a surrogate marker for energetic disruption in ischemic tissue [87, 88]. Cytosolic pH drop causes intracellular Na+ accumulation, which subsequently increases Ca2+ by the Na+/Ca2+ exchanger [89, 90]. Calcium overload contributes to cell death. Under normal physiological conditions, tissue pH is relatively uniform and stable; however, abnormal glucose and oxygen metabolism changes the tissue pH during acute ischemic stroke. Indeed, pH is one of the last indices to change before ischemic tissue progresses towards infarction [28, 91]. A set of optical imaging studies have shown that acidic foci may recruit the ischemic penumbra into infarction [92]. Regi et al. showed that regional cerebral and cortical blood flows are not different between the permanent stroke (60 min) group and another transient ischemia group of four repeated ischemic episodes, 15 min each separated by 5-min reperfusion. At the same time, pH was significantly different [93]. This observation suggests that pH had greater power to define different severity of ischemic tissue injury. However, pH imaging is either invasive [94,95,96,97,98] or of insufficient spatiotemporal resolution for the acute stroke setting [99,100,101].

Amide Proton Transfer pH-Sensitive MRI

Amide proton transfer (APT) imaging, a specific form of chemical exchange saturation transfer (CEST) MRI, has been developed for pH imaging by sensitizing to pH-dependent chemical exchanges between amide protons from endogenous mobile proteins/peptides and water protons [102,103,104,105,106,107,108,109,110]. The endogenous amide proton chemical exchange is dominantly base-catalyzed in the brain, and APT MRI can be used as a sensitive non-invasive pH measurement technique within the physiologically relevant pH [111,112,113]. pH imaging may help define the ischemic penumbra where hypoperfused tissue with intact pH corresponds to benign oligemia. In contrast, the hypoperfused tissue with pH drop identifies metabolic penumbra that is at risk of infarction. The use of pH mapping enables refining the perfusion/diffusion lesion mismatch into acidosis-based penumbra (concurrent perfusion and pH drop) and benign oligemia (hypoperfused tissue with little pH change) [114, 115]. Such additional pH-sensitive metabolic imaging along with the diffusion and perfusion MRI may improve prediction of tissue outcome and ultimately help guide stroke treatment [116].

Substantial progress has been achieved in fast pH MRI to make it amenable in the acute stroke setting. It has been shown that an intermediate RF saturation power level maximizes the pH contrast between the ischemic lesion and the contralateral normal area [117]. The pH-dependent APT MRI effect correlates with lactate concentration, as expected [118]. However, the commonly used pH-weighted image is susceptible to concomitant relaxation and magnetization transfer (MT) that are not pH specific. Zhou et al. showed that T1-normalized CEST effects for the intact white and gray matters are about equal for pH-dependent APT contrast. Yet, MT contrast asymmetry and nuclear overhauser enhancement (NOE) effects are significantly different [119]. Manual lesion outlining has been chosen to overcome pH-weighted image heterogeneity not specific to pH for segmenting ischemic lesions in pH-weighted images [114]. To correct the pH-independent background heterogeneity, Guo et al. postulated that MTRasym (Fig. 3a) heterogeneity from the intact tissue could be described as a multilinear regression (Fig. 3b) of magnetization transfer ratio (MTR) and relaxation (i.e., 1/T1), and dubbed it magnetization transfer and relaxation-normalized APT (MRAPT) analysis [120]. The development of MRAPT analysis improves pH imaging specificity and enables absolute pH mapping [121]. Perfusion and diffusion images (Fig. 3d and f) reveal pronounced perfusion, pH, and diffusion lesion mismatch, which have been postulated to correspond to infarction (diffusion lesion—black, Fig. 3f), metabolic penumbra (pH/diffusion mismatch—green, Fig. 3f), and benign oligemia (perfusion/pH mismatch—red, Fig. 3f). It is worth noting that the diffusion lesion has a worse pH drop than the penumbra [122]. Also, kurtosis lesion suffers from worse pH drop than that of the diffusion/kurtosis lesion mismatch, corroborating the refined mismatch paradigm [123]. Also, the pH-specific MRI allows fast field inhomogeneity correction that minimizes the acquisition time, which makes it highly amenable to the acute stroke setting [124, 125].

Fig. 3
figure 3

Demonstration of pH-specific MRI in an acute stroke rat. a pH-weighted MTRasym. b The multilinear regression between MTRasym, R1w, and MMTR from the intact tissue, per pixel, correction of which results in c pH-specific ΔMRAPTR map. Rat perfusion (d) and diffusion (e) images reveal perfusion/pH/diffusion lesion mismatch (Guo et al., Neuroimage 2016:242–9)

Full size image

Wang et al. demonstrated the potential use of pH lesion in refining acute ischemic tissue in acute stroke rat models [121]. Figure 4 overlays multiparametric MRI indices from the diffusion lesion (black triangle), pH/diffusion lesion mismatch (green square), perfusion/pH lesion mismatch (red circle), and perfusion/diffusion lesion mismatch (solid pink circle). Figure 4a shows that although diffusion lesion has significantly reduced ADC from all three mismatch regions (i.e., pH/diffusion, perfusion/pH, and perfusion/diffusion mismatches), the mismatch regions have substantially overlapped perfusion and diffusion values. Although all ischemic areas have significantly reduced cerebral blood flow (CBF) from the contralateral brain, only PWI/pH mismatch has a significantly higher perfusion level than the diffusion lesion (Fig. 4b). Figure 4c shows that while ADC cannot differentiate perfusion/diffusion, perfusion/pH, and pH/diffusion mismatches, their pH was significantly different, being 6.84 ± 0.10, 7.01 ± 0.04, and 6.71 ± 0.12, respectively. This data suggests the potential use of pH to sensitize the heterogeneous metabolic disruption within the routine PWI/DWI lesion mismatch. Indeed, Fig. 4d shows that regions of diffusion lesion, pH/DWI lesion mismatch, and PWI/pH lesion mismatch can be clustered using multi-dimensional perfusion, pH, and diffusion indices, augmenting routine perfusion and diffusion-based stroke imaging.

Fig. 4
figure 4

Comparison of perfusion, pH and diffusion indices from diffusion lesion, pH/diffusion lesion mismatch, perfusion/pH lesion mismatch, and perfusion/diffusion mismatch from all animals. a ADC vs. CBF. b pH vs. CBF. c pH vs. ADC. d Three-dimensional stratification of CBF, ADC, and pH indices from diffusion lesion, pH/diffusion lesion mismatch, and perfusion/pH lesion mismatch. CBF, cerebral blood flow; ADC, apparent diffusion coefficient

Full size image

It is worth noting that there could be non-negligible T1 changes following the acute stroke [126,127,128], while the mean MT ratio (MMTR) from ± 3.5 ppm shows little change [121]. After accounting for the difference in relaxation time, a recent study showed that the pH-sensitive APT signal dominates the NOE effect, supporting the continued use of MTRasym and amalgamations of it (e.g., MRAPT analysis) for pH imaging in the acute stroke setting [129].

New Imaging-Based Ischemic Tissue Classification

The recent development of DKI and pH MRI provides a tangible means to refine the mismatch paradigm (Fig. 5). Figure 5a shows the routine PWI/DWI mismatch paradigm, in which the DWI lesion defines the infarction core (black), and the PWI/DWI lesion mismatch identifies the salvageable penumbra tissue (red). With the development of DKI and pH imaging, we postulate that the heterogeneous ischemic lesion can be refined (Fig. 5b). Specifically, the PWI/DWI mismatch includes benign oligemia (hypoperfusion tissue without pH change, region IIb in red) and metabolic penumbra (hypoperfused acidic tissue, region IIa in light green). In addition, the DWI lesion contains an irreversible infarction core (DKI lesion, region Ia in black) and a portion of DWI lesion that is still salvageable despite its worse pH drop than region IIa (DWI/DKI lesion mismatch, region Ib in dark green). Altogether, the penumbra is defined by pH/DKI lesion mismatch (regions Ib (dark green) + region IIa (light green)). Although additional work is needed to fully establish that pH/DKI mismatch is the penumbra, accumulating data suggest that advanced stroke imaging has the potential to augment the routine perfusion/diffusion mismatch paradigm. A fast and refined tissue characterization may provide the urgently needed imaging evidence to individualize and transform the state-of-the-art stroke patient care.

Fig. 5
figure 5

Illustration of the refined mismatch paradigm. a The routine PWI/DWI mismatch paradigm. b Modified penumbra defined by pH/DKI mismatch

Full size image

Clinical Translation of DKI and pH Imaging

Although it may seem challenging to incorporate advanced MRI to the acute stroke clinical settings, there have been increasing reports of DKI and pH stroke imaging, particularly in large stroke centers where the infrastructure and workflow enable exploration of novel stroke imaging and treatment. Specifically, pH-sensitive APT MRI has been translated to study acute stroke patients with preliminary yet promising results, especially for those with relatively delayed presentation of 24–48 h [130,131,132,133]. For example, Heo et al. demonstrated pH-sensitive imaging in acute stroke patients at 3 T, with a scan time of 3 min 14 s [133]. Because the simplistic MTRasym index includes contributions from APT, NOE, and possibly relaxation changes, the APT effect was quantified using an extrapolated semisolid MT reference signal technique (rAPT#). The pH/diffusion and perfusion/pH scatter plots from two representative stroke patients are shown in Fig. 6. The first patient had a perfusion/pH lesion mismatch with minor pH/diffusion mismatch (Fig. 6a). The scatter plot showed two clusters; the diffusion lesion showed significantly relative ADC (rADC) drop (Fig. 6b), while the perfusion/pH lesion mismatch showed intact rADC and delayed relative bolus transit time (rBTT), as expected (Fig. 6c). In the second stroke patient, there were both noticeable pH/diffusion and perfusion/pH lesion mismatches (Fig. 6d). The rADC and rAPT# scatter plots show that the pH/diffusion lesion mismatch is of higher rADC than the diffusion lesion (Fig. 6d). Although pH/diffusion lesion mismatch has lower rAPT# than the contralateral normal tissue, the pH/diffusion lesion mismatch and diffusion lesion had reasonably overlapped pH-weighted MRI contrast. Also, pH/diffusion and perfusion/pH lesion mismatches displayed delayed bolus transit time (Fig. 6f).

Fig. 6
figure 6

Comparisons of diffusion/pH/perfusion deficits, pH/diffusion, and perfusion/pH scatterplots in two acute stroke patients at 1 day from symptom onset. a–c A patient with pH/perfusion mismatch, but minor diffusion/pH mismatch. d–f A patient with pH/perfusion mismatch, as well as diffusion/pH mismatch. The distributions of the diffusion deficit area (red), pH–diffusion mismatch (green), and perfusion–pH mismatch (blue) were markedly different from those of the contralateral normal tissue (black)

Full size image

For the clinical translation of DKI, Yin et al. implemented a multi-band fast DKI protocol at 3 T, with a scan time of 2 min 10 s. They reported that for ischemic lesions over 1 cm in diameter, kurtosis lesions are of stronger correlation with the follow-up T2 MRI than those of diffusion MRI [134]. Also, Zhu et al. studied 156 stroke patients and analyzed 199 lesions in regions of periventricular white matter, corpus callosum, cerebellum, basal ganglia and thalamus, brainstem, and gray–white matter junctions. They concluded that DKI could reveal the differences in microstructure changes among various locations affected by acute ischemic stroke and performed better than diffusivity [135]. Guo et al. compared MK, axial kurtosis, and radial kurtosis in acute stroke patients. They concluded that axial kurtosis is better suited for diagnosing acute ischemic lesions in highly anisotropic brain regions, such as the corpus callosum and corona radiate. In contrast, MK may be appropriate for the lesions in low anisotropic or isotropic brain regions, such as the thalamus, subcortical white matter, and cerebral cortices [136].

Future Directions for Acute Stroke Imaging and Acute Stroke Research

Advanced imaging may help guide the development of new stroke therapeutics such as novel neuroprotection in combination with effective reperfusion like alkalinizing agents [137], Na+/H+ exchanger (NHE1) [138, 139], and acid-sensing ion channel (ASIC) blockers [140, 141]. In particular, the most recent Stroke Preclinical Assessment Network (SPAN) aims to test new compounds/interventions in animal models of cerebral ischemia following the stroke treatment academic industry roundtable recommendations [142,143,144,145]. Future work to reduce the scan time for DKI and pH imaging is needed to enable a full panel of stroke MRI examinations without delaying the treatment. Advanced stroke imaging could also facilitate their translation and benefit new acute stroke trials, building on the successes of DAWN and DEFUSE-3.