Ischemic brain tissue damage mechanisms are complex, and have been categorized into pathophysiology, biochemistry, gene expression, cell signaling, pharmacology, and neuroimaging changes [1-3]. Magnetic resonance (MR)-based techniques have been playing a vital role not only to assist our understanding of stroke pathology but also to help guide stroke patient management [4-10]. As such, it is important to define relevant pathophysiologic mechanisms in stroke as determined by molecular and functional neuroimaging, identify and prospectively validate surrogate imaging biomarkers of tissue injury for early prediction of tissue outcome, and ultimately guide individualized stroke treatment, both in experimental stroke models and translational clinical investigation [11-13]. Our article provides a concise overview of stroke pathology, conventional and emerging MRI and MRS techniques, and multi-parametric data analysis tools that characterize the hemodynamic, metabolic, structural impairments subsequent to ischemic injury.

Ischemic stroke occurs when a major cerebral artery becomes blocked, causing severe hypoperfusion to the brain. Blood clots are the most common cause of artery blockage and brain infarction. After ischemic stroke brain tissue ceases to function and latterly suffers irreversible injury possibly leading to death of the tissue, i.e., infarction. Upon ischemia, tissue becomes low in energy substrates including glucose and oxygen, and resorts to anaerobic respiration. Anaerobic glycolysis, however, produces less adenosine triphosphate (ATP) yet releases lactic acid as a byproduct. Lactic acid is an irritant and at high concentration, could further disrupt tissue metabolism and potentially destroy cells [1]. The major cascade of neuronal injury and brain infarction after ischemic stroke is initiated when the energy production such as ATP fails, leading to failure of energy-dependent processes (such as ion pumping) that are vital for cell viability [14]. Another major cause of neuronal injury is the release of excitatory neurotransmitters such as glutamate. The extracellular glutamate concentration is normally kept low, powered by the concentration gradients of ions (mainly Na⁺) across the neuronal cell membrane. After stroke, the energy-dependent trans-membrane ion gradients run down, and glutamate transporters reverse their direction, releasing glutamate into the extracellular space. Glutamate acts on receptors in neuronal cells and therefore, produces an influx of calcium that activates enzymes that digest cellular proteins, lipids and nuclear materials. Calcium influx can also lead to the failure of mitochondria, which may cause further energy deterioration. Ischemia also induces production of free radicals and other reactive oxygen species, which may react with and subsequently damage a number of cellular and extracellular elements. In addition to injurious effects on brain cells, the loss of neurovascular structural integrity and functional coupling may also result in neurological dysfunction, a breakdown of the protective blood brain barrier (BBB) that contributes to cerebral edema, which can cause secondary progression of the brain injury [3, 15].

Emerging evidence suggests that ischemic brain damage is a dynamic process that evolves over time. The progression and the extent of ischemic injury are influenced by many factors, including age, severity and location of occlusion, state of collateral and systemic circulation, hematological and coagulation factors, body temperature and blood glucose level, etc [16]. Furthermore, acute ischemic tissue often suffers a gradient loss of perfusion, rather than complete and homogeneous ischemia of the entire occluded vessel supplying territory [10]. The key pathophysiological concept is the delineation of ischemic tissue into three operational compartments. Regions suffering the most severe hypoperfusion rapidly progress to irreversible damage and inevitably infarct, representing the ischemic core. The remaining hypoperfused tissue exhibits impairment of the normal blood flow autoregulatory mechanisms and can be pathophysiologically divided into two compartments, namely, salvageable tissue at risk of infarction ‘penumbra’ and mild ischemic tissue at no immediate risk of infarction ‘oligaemia’[17]. The ischemic penumbra was originally described on electrophysiological basis as the tissue existing between the thresholds of electrical failure and ion pump failure [18]. In the ischemic penumbra, oxygen metabolism is preserved and thereby potentially salvageable. Its extent however, decreases over time due to gradual deterioration and as such, represents a key target for therapeutic intervention [19]. Whereas this course of events varies from patient to patient, it has been shown that ischemic penumbra may last for many hours after stroke onset [20]. Timely rescue of the penumbra, either by restoration of blood supply or interruption of the adverse metabolic or neurochemical cascades, is the basis of acute stroke therapy. The benign oligaemic tissue, on the other hand, suffers a mild degree of hypoperfusion with normal oxygen consumption and elevated cerebral blood volume (CBV) and oxygen extraction fraction (OEF), and is not at immediate risk of infarction. If the occlusion persists, however, secondary events such as systemic hypotension, intracranial hypertension or hyperglycaemia may topple this delicate balance and induce the oligaemic tissue to transform into penumbral state and eventually being recruited into the necrotic core [17].

Magnetic resonance spectroscopy (MRS) is capable of studying cellular biochemistry and metabolism by measuring various cerebral metabolites, providing a noninvasive and sensitive means to assess acute ischemic stroke and its progression [21]. Studies of metabolites using proton (¹H) and phosphorus-31 (³¹P) MRS have enormously contributed to our understanding of cellular metabolism and pathophysiologic processes of cerebral ischemia. Single-voxel based MRS techniques such as point resolved spectroscopy (PRESS) or stimulated echo acquisition mode (STEAM) have been increasingly employed in stroke exams due to its widespread availability and reasonable acquisition time. Recently, there has been great interest in the development of magnetic resonance spectroscopic imaging (MRSI), which can provide information regarding the spatial distribution of metabolites after stroke [22-26] by means of metabolite maps [25], despite moderately longer acquisition time.

Among various metabolites detectable by ¹H MRS, increase in lactate (1.33 ppm) is the most prominent feature of ischemic stroke. Lactate, produced as a metabolic end-product of anaerobic glycolysis after the onset of ischemic insult, provides a sensitive means for detecting acute stroke [27, 28]. In ¹H MR spectra obtained at short echo time, lactate signal overlaps with lipid peaks (0.9-1.4 ppm), rendering quantitative detection of lactate difficult. Due to the short T₂ relaxation time of lipids, long echo time spectra can suppress the lipid signal and facilitate the measurement of lactate [29]. It is noteworthy that the signal of lactate modulates with echo time due to J-coupling [30], and the lactate doublet peaks are in phase and usually acquired at TE = 136 (inverted) or 272 ms. Apart from lactate, N-acetyl aspartate (NAA; 2.01 ppm) is found to decrease after onset of stroke and continues to decline with time in ischemic lesion [27, 28]. Being highly correlated with histological findings in animal studies, NAA signal decrease indicates neuronal loss or damage after cerebral ischemia [31, 32]. As such, NAA may serve as a metabolic marker in determining early tissue viability. The degree of ischemia and neuronal viability as determined by levels of lactate and NAA, respectively, during acute phase of ischemic stroke has been shown to correlate with the severity of ischemic insult and outcome [33, 34]. Maniega et al. demonstrated that temporal evolutions of both lactate and NAA concentrations in stroke patients can provide useful insights into the dynamics of ischemic stroke [35]. Furthermore, it has been suggested that region with near normal NAA but elevated lactate level may represent salvageable ischemic penumbra in acute stroke [25, 36, 37].

³¹P MRS has been used to study energy states in ischemic stroke by assessing the high-energy phosphorus-containing moieties participating in energy metabolism, particularly ATP and phosphocreatine (PCr) [38-42]. During ischemia, PCr energy buffer decreases with the increase of inorganic phosphate (P_i) in order to maintain ATP homeostasis, and ATP levels decreases once PCr buffer is depleted. In addition, ³¹P MRS can provide information about intracellular acidosis by determining the difference in chemical shift between the P_i and PCr peaks (δ) as pH = 6.72 + log((δ - 3.27) / (5.69 – δ)) [38, 39, 41, 43].

Commonly used stroke MRI methods include perfusion, diffusion and relaxation MRI. In addition, magnetization transfer (MT) and pH-weighted amide proton chemical exchange saturation transfer (CEST) MRI are also being explored for stroke imaging. It is important to note that whereas computed tomography (CT) is the most utilized method while positron emission tomography (PET) provides more specific characterization of tissue metabolism and perfusion, MRI is widely used due to its multi-parametric diagnosis capability, relatively easy access and non-ionization radiation [8, 11, 13, 44-46].

Due to the dynamic nature of ischemic tissue damage, a single MRI parameter often has its own time window to reveal the abnormality, and thus may not fully characterize the ischemic tissue injury [133]. For instance, ADC declines rapidly within minutes of ischemic stroke, but it pseudo-normalizes and even rises above the normal range in latter stages [68]. Therefore, the multi-parametric MRI analysis is necessary to assess the heterogeneous ischemic tissue damage [13, 134]. Specifically, methods including generalized linear model (GLM), K-means, fuzzy c-means, and interactive self-organizing data analysis technique algorithm (ISODATA) segmentation with multi-parametric MRI have been suggested for ischemic tissue classification [44, 135, 136]. In particular, ISODATA is an unsupervised segmentation algorithm based on cluster analysis which can recognize structures within a data set and automatically determine the number of clusters [137]. Multiple tissue signatures can be segmented from the data set by ISODATA, generating a theme map that reflects different tissue clusters [138]. Studies have shown that ISODATA technique can accurately define the ischemic region which was well correlated with histologically determined lesion at different stages after stroke [136, 139], and ISODATA lesion volume at acute stage has strong correlation with stroke outcome [138, 140]. Increasing the dimensionality of the ISODATA model by incorporating additional images has been shown to better demarcate tissue clusters [139, 141]. Moreover, the heterogeneity within the ischemic lesion in the ISODATA theme map has been suggested to indicate heterogeneous tissue damage [142]. Furthermore, Shen et al. showed that employing statistical algorithm following ISODATA tissue clustering can help improve the prediction of the ischemic tissue outcome in animal models [143, 144]. Specifically, a training data set was used to derive the probability profiles of tissue fate pixel-by-pixel, which were then applied to generate maps of risk of subsequent infarction. Such a predictive approach has been shown to improve the predication in transient stroke by taking into account of the regional susceptibility to infarction [144]. More recently, predictive algorithm based on artificial neural network has also been developed for prediction of ischemic tissue fate based on multiparametric MRI data [145]. These predictive models, if fully developed, may greatly aid clinical decision making in the treatment of acute stroke by providing objective predictions of ischemic tissue fates.

MRI and MRS techniques have greatly improved our understanding of stroke pathophysiology. In addition, emerging MR techniques are being developed to capture new facets of ischemic tissue damage with enhanced sensitivity and specificity. Such a breadth of information can be characterized with multi-parametric analysis tools for improved tissue classification. Most importantly, further development and validation of MR techniques and image analysis tools may help establish imaging-based outcome prediction algorithms and ultimately, guide individualized stroke treatment.