The imaging evaluation of patients suspected of acute ischemic stroke has markedly changed over the past few years. It is no longer limited to unenhanced head CT but instead includes vascular imaging as a standard and increasingly tends to integrate perfusion imaging. Therefore, it is important for radiologists to be familiar with multimodal CT, including its strengths and limitations. The 10 questions discussed in this article represent the most frequent questions that we receive from colleagues interested in building a stroke program involving perfusion CT imaging.

In the assessment of acute cerebral ischemia, perfusion CT and MRI provide similar information. The infarct core and the ischemic penumbra as shown by perfusion CT and by diffusion-weighted imaging (DWI) and perfusion-weighted imaging (PWI) are comparable [1, 2]. Nonetheless, despite similarities in terms of their results, CT and MRI techniques have their respective technical advantages and limitations.

MRI offers whole-brain coverage and does not involve x-rays and is thus preferable in younger patients because of a reduced amount of lifetime radiation exposure. Furthermore, it can be performed without administration of contrast material (e.g., time-of-flight MR angiography [MRA] and arterial spin labeling perfusion imaging), and hence is appropriate even in patients with renal failure or allergy to contrast material. MRA and perfusion-weighted MRI can also be performed using gadolinium contrast media; however, MR contrast material is contraindicated in patients with renal failure because of the risk of nephrogenic systemic fibrosis [3]. MRI has a number of absolute contraindications. Patients with mechanical implants, electronically operated devices, or ferromagnetic hemostatic clips in the brain should not be examined with this technique. MRI is usually contraindicated in patients with cardiac pacemakers; nonetheless, it can be performed in some selected patients with certain types of pacemakers using specific safety precautions [4].

CT uses x-rays and iodinated contrast material. Yet, when using the correct acquisition parameters (80 kVp and 100 mAs), the effective radiation dose associated with a perfusion CT study is similar to that of a conventional head CT—around 2 mSv (see answer to question 10). The renal safety of multimodal CT in acute stroke patients was assessed in a large series of 198 patients with suspected acute ischemic or hemorrhagic stroke who underwent perfusion CT and/or CT angiography (CTA), without knowing the baseline creatinine level [5]. None of the patients developed chronic disease or required dialysis. Five patients developed contrast-induced nephropathy (defined as a 25% or more increase in baseline creatinine levels within 72 hours of contrast administration). Fifty-five patients (36.6%) underwent repeat perfusion CT and CTA, additional CTA, or additional cerebral angiography within 24 hours of the initial perfusion CT and CTA, with a total of around 250 mL of injected contrast medium. None of these patients with additional contrast studies developed contrast-induced nephropathy or chronic kidney disease despite the higher volumes of contrast material used. Thereby, perfusion CT was shown to be safe even without measuring the serum creatinine level before injection of iodinated contrast material, provided that the patient does not have a history of renal failure [5].

In its earliest stages of use, perfusion CT was limited in spatial coverage to an imaging slab of 20 mm (for 16-MDCT scanners) or 40 mm thickness (64-MDCT scanners). This issue has been addressed on modern CT scanners, either through the development of greater arrays of CT detectors or using different variants of the toggling table technique [6]. Using the full width of the 320-MDCT scanner's capability enables full brain coverage in a single CT gantry rotation and provides the ability to obtain combined volumetric perfusion and angiographic data from the same, single contrast material injection [7]. Increased z-axis coverage can also be obtained for each bolus by using a toggling table technique, in which the CT scanner table moves back and forth, alternating between two different locations [6]. This technique offers the double advantage of increased spatial coverage and lower radiation dose. However, it has the limitation of diminished temporal resolution, which may in turn affect the quantitative accuracy of the perfusion CT results.

One advantage of perfusion CT relative to MR perfusion imaging relates to the superior quantitative accuracy offered by perfusion CT. MR perfusion imaging affords only semiquantitative comparison of one hemisphere to the other. However, the quantitative accuracy of perfusion CT, which has been shown by comparison with both xenon-CT and PET [8, 9], allows the use of absolute perfusion thresholds as well as the monitoring of brain perfusion over time; the latter is useful for assessing the efficacy of a therapeutic intervention. Nonetheless, the key element supporting the use of CT in the initial evaluation of patients suspected of acute ischemic stroke is its wide availability, which is especially important in the emergency setting. CT scanners are available throughout the day and night in the wide majority of medical institutions. Perfusion CT and CTA have low technologic requirements and a short imaging time (typically 10–15 minutes). On the other hand, emergency round-the-clock MRI access is still available only in a limited number of hospitals and thus is usually reserved for specific stroke patients (see question 2).

Overall, the combination of CTA and perfusion CT and MRA and MR perfusion are considered to be equivalent tools in the evaluation of patients with suspected acute ischemic stroke [1, 2]. However, there are some specific clinical settings in which one technique is preferred. For instance, MRI is superior to CT for detection of small (lacunar) infarcts and posterior fossa ischemic lesions; thus, it would be preferred in that clinical setting [10]. However, when the primary issue is that of arterial patency, CTA should be the first-line diagnostic test [11]. Current CT scanners quickly (within 5 seconds) provide a detailed evaluation of the intracranial and extracranial vasculature [12, 13] in a single data acquisition with excellent isotropic spatial resolution. Multiplanar reformatted images, maximum intensity projection (MIP) images, and 3D reconstructions of source images provide detail comparable to that obtained with conventional (catheter) angiography [12, 14]. CTA is clearly superior to MRA in evaluating both the intracranial and extracranial circulation and less prone to artifacts, resulting in more accurate quantification of stenoses [11].

When evaluating the patient with acute cerebral infarction who is in the therapeutic window for thrombolytic therapy, detection of the presence of hemorrhage is important because it has important ramifications for treatment with thrombolytic agents. Because an answer must be obtained quickly, CT is the preferred technique for this purpose. It is generally agreed that MRI with gradient-echo or susceptibility weighted sequences is more sensitive than unenhanced head CT for detection of microbleeds [15, 16], but such hemorrhages have not been clearly shown to be a contraindication to thrombolysis [17], making this advantage of MRI less important.

A number of medical centers perform CTA before perfusion CT, using the rationale that CTA is thought to provide more vital information compared with perfusion CT and should be obtained before patient motion (which is often more substantial later in the scan period) interferes with data acquisition. Thus, performing CTA first maximizes the chances that the more important CTA data are obtained without degradation by motion. In addition, performing CTA first theoretically avoids the venous contamination from an initial perfusion CT contrast injection and, thus, may improve arterial lesion detection and characterization. Furthermore, the area showing a paucity of vessels on CTA source images has been shown to correlate with the infarct core [13, 18]. Performing the CTA study first allows characterization of the exact anatomic location of the ischemic area. The radiologist or technologist can then place the perfusion CT slices in the area of suspicion, which formerly was a concern when the spatial coverage of perfusion CT was limited.

At our institution, patients with clinical suspicion of acute stroke with no contraindications to IV iodinated contrast material undergo a stroke imaging evaluation that includes (in the following specific temporal sequence) an unenhanced head CT, perfusion CT using 40-mm slabs at two locations, and CTA of the extracranial and intracranial arteries. Perfusion CT source images are used as a test bolus to determine the optimal timing of CTA (Fig. 1). In our protocol, acquisition of perfusion CT images starts 7 seconds after the beginning of the injection of IV contrast material (40 mL of contrast material injected at 5 mL/s). After first performing the perfusion CT series, the CT technologist reviews the perfusion CT images and identifies the image frame where the contrast material starts appearing in the arteries. The time of this image frame (e.g., image frame obtained 13 seconds after the beginning of the perfusion CT acquisition) is added to the starting time of the so-called prep group (7 seconds), and the resulting time delay (20 seconds in this example) is used as the delay between the beginning of the IV injection of contrast material for the CTA (50–70 mL of contrast material injected at 5 mL/s) and the acquisition of CTA images (contrast injection, of course, needs to remain exactly the same). This process ensures an optimal arterial phase of the CTA because very little residual venous contrast material remains from the perfusion CT acquisition in the presence of adequate renal excretion. Our technique is illustrated in Figure 1.

One negative aspect of performing perfusion CT before CTA is that some degree of venous contrast enhancement (caused by perfusion CT contrast material administration) will be present on CTA images. However, in our experience, the advantages provided by optimal timing of CTA imaging (allowed by the prior perfusion CT contrast administration) outweigh the disadvantages of perfusion CT–related venous contamination of the CTA images. Another fact supporting performance of perfusion CT before CTA is that the relatively large amount of contrast material used for CTA (i.e., 50–70 mL) causes some baseline contamination that interferes with calculation of accurate perfusion values on perfusion CT maps. In other words, venous contamination is more of a problem for perfusion CT after CTA than for CTA after perfusion CT.

In the future, the use of volumetric full-brain coverage CT scanners or the wider availability of improved large coverage toggling table acquisition techniques will allow combined volumetric perfusion and angiographic data from a single contrast injection and render the considerations outlined earlier irrelevant [7].

The primary goal of acute stroke imaging is to provide an assessment of ischemic tissue viability on the basis of the relative extent of infarcted tissue (region of tissue that cannot be salvaged) and penumbra (region that is at risk for infarction in the absence of adequate therapy).

Perfusion CT provides information on three perfusion parameters: mean transit time (MTT), cerebral blood flow (CBF), and cerebral blood volume (CBV). MTT designates the average time required by a bolus of blood to cross the capillary network. CBF is a measure of the volume of blood flowing per unit of brain volume during a time interval of 1 minute [19, 20]. Finally, CBV refers to the blood volume per unit of brain volume. The relationship between CBV, MTT, and CBF is expressed by the equation CBF = CBV / MTT [19, 20]. Thus, once two hemodynamic parameters are known, the third can be calculated by using the central volume principle.

Perfusion CT in patients with acute cerebral ischemia provides the means to distinguish infarct core from penumbra. To understand how this distinction is made by perfusion CT, it is worth reviewing the concept of cerebral vascular autoregulation [1, 19, 21]. Vascular autoregulation is the mechanism by which regional CBF is maintained despite changes in the metabolic activity of local neurons and in the local arterial perfusion pressure. Autoregulation is controlled by complex neurobiochemical mechanisms involving sensitivity to blood pressure, blood pCO2, and pH. Cerebral ischemia is characterized by decreased perfusion pressure, which produces prolongation of the MTT in both the ischemic core and the penumbra; in response, exclusively in the penumbral tissue and not in the infarct core, the process of autoregulation induces dilation of capillaries supplying and draining the ischemic region in an attempt to maintain a constant CBF. As a result, CBV in the penumbral tissue is maintained or increased due to vasodilatation (Table 1). However, in the infarct core, autoregulation mechanisms are overwhelmed and lost, and CBV is diminished [1, 19] (Figs. 2 and 3). Thus, the perfusion CT parameter that most accurately describes penumbral tissue in the first few hours after onset of ischemia is the relative MTT; we have found an MTT threshold of 145% (elevated MTT of at least 145% compared with normal contralateral tissue) to be optimal as a means of defining tissue with prolonged MTT. The parameter that most accurately describes the infarct core is the absolute CBV, with an optimal threshold of 2.0 mL/100 g. Thus, the tissue in which MTT is elevated contains both the penumbral tissue and the ischemic core; the penumbral tissue can be identified on a perfusion CT map as that tissue in which a mismatch between MTT and CBV is present, i.e., in which relative MTT is elevated above the threshold value but absolute CBV is not decreased below the threshold value [22] (Fig. 3).

After perfusion CT data acquisition, source images are transferred to a postprocessing workstation where dedicated software allows creation of parametric maps for clinical interpretation. Like all images, perfusion CT maps are composed of pixels. For each pixel on a perfusion CT map, hemodynamic parameters are calculated on the basis of the time–density curves after infusion of contrast material, reflecting the wash-in and wash-out of contrast material in this pixel. The CBV calculation from perfusion CT studies relies on the assumption that the perfusion tracer (e.g., contrast material) is not diffusible into the brain tissue that it traverses (i.e., is confined to the vascular compartment). If this assumption is indeed true, then the time–density curves observed in pixels within vessels should have a different appearance (different area under the curve) from those pixels within brain parenchyma. This is indeed the case; the area under the curve is greater in voxels containing solely or predominantly vessels than for voxels containing a mixture of brain tissue and capillaries (for which the vascular volume represents only a few percent of the total volume). Determination of the contrast enhancement profile in a reference pixel that contains solely blood (e.g., at the center of a large vein) allows calculation of the fraction of vascular volume within pixels that have a mixture of blood and brain tissue (pixels within brain parenchyma). This technique is thus based on calculating the amount of partial volume averaging effect. In this manner, the relative CBV of brain tissue can be calculated on a pixel-by-pixel basis and a CBV map can be derived [19, 23]. The selection of the appropriate vein for this calculation is discussed in question 7.

The vascular confinement assumption is applicable to healthy brain tissue. However, when substantial blood–brain barrier breakdown is present (e.g., contrast-enhancing inflammatory or neoplastic lesions), leakage of contrast material into the extravascular space causes overestimation of the fractional vascular volume and thus of the CBV [24]. As a useful rule, the presence of contrast enhancement on standard CT images can be used to determine whether overt leakage of contrast material is present.

As opposed to the methods for determination of CBV outlined here, commercial software suppliers use one of a number of different mathematic models to calculate the MTT and CBF. The two major models are the maximum slope model and deconvolution-based mathematic methods (which are attaining more widespread acceptance) [19, 23]. These models are outlined next.

The maximum slope model was initially described for intraarterial injection of colored or radiolabeled microspheres that were completely extracted at the first pass through a tissue [19]. According to the Fick principle, the perfusion of a particular territory is proportional to the total number of extracted particles in this territory and to the rate of accumulation of the microsphere (slope of the accumulation curve). The maximum slope model can be applied to perfusion CT studies; the CBF in a voxel of brain tissue during a bolus injection is considered to be reflected by the maximum slope of density increase on the time–density curve recorded in that voxel [19, 23]. A limitation of application of this model to cerebral perfusion is that iodinated contrast material is not extracted during the first pass. As a compensatory measure to allow use of this model, one may assume that during the time period of infusion, the venous concentration is zero (no contrast material has yet reached the venous side of the circulation). However, this condition would require a very high injection rate of contrast material (at least 10 mL/s) that is not realistically attainable in most patients (in whom 5 mL/s seems to be the maximum bearable injection rate in the routine setting of a small peripheral IV line) [19]. If the injection rate is too low, fractions of the bolus of contrast material are not accounted for and, as a result, CBF values are underestimated [19]. Thus, the maximum slope model suffers from two limitations, i.e., inappropriate assumptions of the absence of a contribution of venous concentration and the potential for underestimation of the CBF.

The deconvolution method provides a means to overcome the limitations of the maximum slope method because it does not rely on simplified assumptions regarding the underlying vascular architecture and is reliable even for low rates of injection of contrast material [25]. This method consists of calculating MTT or CBF from the shape of the time–density curves of the tracer at the arterial input and at the brain tissue of interest [19, 23]. The tissue time–density curve represents a combination of the effects of the arterial input function and the inherent tissue properties. The choice of the arterial input function directly affects the calculation of MTT and CBF but not CBV; instead CBV is calculated on the basis of the amount of partial volume averaging effect using a reference vein (as outlined earlier) [19, 23]. Before calculating MTT and CBF, the effects of the arterial input function on the tissue concentration curve must be removed by using a mathematic process known as “deconvolution,” which allows the true hemodynamic properties of the voxel under consideration to be derived [19, 23]. Early versions of the deconvolution method were technically demanding and involved complicated and time-consuming data processing, which delayed widespread clinical application [19, 23, 25]. However, currently available commercial software allows generation of MTT and CBF parametric maps quickly (i.e., in less than 1 minute) and easily; they can thus be used in the emergency setting of acute cerebral ischemia.

The original model recommended selecting the arterial input function as close as possible to the arterial territory being evaluated (e.g., in the case of middle cerebral artery ischemia, an artery that supplies that territory). Thus, the arterial input function is estimated from a major artery, with the assumption that it represents the exact input to the tissue, i.e., that no significant delay exists between that artery and the region of tissue being examined. However, this assumption is often not valid because the reference artery is at some distance from the tissue of interest. Thus, any delay of the bolus during its passage from the reference artery to the tissue of interest may introduce errors in quantification of CBF or MTT [26, 27].

Several studies have shown that, in acute stroke patients, the selection of the arterial input function does not have a significant impact on the perfusion CT parametric maps if the ipsilateral carotid artery is not stenotic [28–30]. Some studies have shown that MTT and CBF values are independent of the type (large proximal vs small distal artery) [28] or laterality (ipsilateral vs contralateral to the infarct location) of the artery chosen as the arterial input function [29]. Thus, in an acute stroke patient, a proximal anterior cerebral artery branch is usually selected as the arterial input function for the sake of simplicity; because they are perpendicular to axial images (which decreases the risk of volume averaging with brain tissue) and have relatively large caliber, they can be easily and reproducibly identified [30].

In patients with known chronic vascular conditions (e.g., chronic internal carotid artery stenosis or giant aneurysm) in which collaterals play a preponderant role and cause delay and dispersion of the contrast bolus, artery selection can have a significant impact on MTT and CBF calculation (Fig. 4). The so-called delay-corrected or delay-sensitive deconvolution software is especially useful in such settings (see question 8). Alternatively, the time–density curve of each vascular territory can be deconvolved using an arterial input function derived from its own specific parent artery [26] (Fig. 4).

As mentioned earlier, CBV values are affected solely by the choice of a reference venous output function (rather than by choice of the arterial input function) [19, 23]. The site for a reference venous output function should be chosen using a voxel having the maximal area under the time–density curve or, stated differently, the least amount of partial volume averaging. The voxel at the center of the superior sagittal sinus, which has the advantage of being large and orthogonal to axial perfusion CT source images, is adequate for this purpose in a large majority of cases. In other instances, other venous structures, such as the straight sinus, also can be used. In fact, the calculation of CBV only requires the selection of a solely vascular pixel (devoid of partial averaging effect) having maximal area under the time–density curve. Thus, at times a large artery within the perfusion CT slice that can confidently be claimed to be devoid of partial volume averaging effect, e.g., the cavernous internal carotid artery, can be chosen for the “venous” output function.

Indeed, as stated in the response to question 6, when using delay-sensitive deconvolution software, a delay in transit of contrast material due to a hemodynamically significant carotid stenosis causes prolonged MTT values in the territory supplied by the stenotic artery, even in the absence of brain ischemia (Fig. 4). This phenomenon reflects not only the delay caused by the stenosis but also by the fact that collateral arteries typically supply such a vascular territory; transit of blood (and contrast material) through collateral arteries is slower because collateral circulation is an indirect route to tissue. Although this phenomenon is the cause of apparently high MTT values (and should not be misinterpreted to represent signs of acute cerebral ischemia), it has clinical relevance and should not be ignored. This delay and the resulting falsely increased MTT likely reflect a decrease in cerebral vascular reserve and an increased likelihood of ischemia in the future. This phenomenon serves as a reminder that when interpreting perfusion CT parametric maps, the radiologist should always review the CTA images to detect possible arterial stenoses or occlusions that may cause delay or dispersion of the bolus of contrast material.

When delay-insensitive or delay-corrected deconvolution perfusion CT software is used or when the CTA is normal but the perfusion CT shows an MTT abnormality in a vascular territory, the possibility of a very recent transient ischemic attack should be considered. The patient may not be symptomatic at the time of perfusion CT but was likely symptomatic shortly beforehand. Finally, if the perfusion CT abnormality does not conform to a vascular territory, the possibility of a stroke mimic should be entertained (see question 9).

Not all perfusion abnormalities seen on perfusion CT parametric maps are specifically related to cerebrovascular disease. Indeed, many neurologic diseases causing symptoms simulating cerebrovascular disease produce an alteration of brain perfusion and thus can result in perfusion CT abnormalities. The unenhanced CT and CTA images that are performed along with the perfusion CT maps allow one to detect findings indicative of entities that may produce nonischemic perfusion abnormalities. Examples of such entities include subdural hematoma, neoplasm, vasospasm, venous thrombosis, or even seizure related hyper- or hypoperfusion.

The clinical presentation of seizures is variable and can sometimes be confused with that of an ischemic stroke. The specific patterns observed on perfusion imaging depend largely on the timing of the study in relation to the seizure activity, i.e., whether ictal, postictal, or interictal. In our experience [31], interictal perfusion imaging shows increased MTT with decreased CBF and CBV. Although these findings are similar to those of ischemic stroke, they may be differentiated by recognizing that the perfusion alterations related to seizure do not correspond to arterial vascular territories. Rather, the perfusion CT abnormalities after a seizure tend to involve the cortical gray matter, while sparing the white matter (Fig. 5). Imaging within 3 hours after seizure termination increases the chance of finding postictal perfusion CT abnormalities, but perfusion CT abnormalities can be seen in some patients up to 24 hours after the ictus.

Cerebral venous thrombosis is a relatively uncommon but serious neurologic disorder that can present with strokelike symptoms. Early diagnosis of cerebral venous thrombosis is important because prompt medical therapy can reverse the disease process and avoid brain injury [32]. Perfusion abnormalities described in patients with venous thrombosis consist of increased MTT with preserved CBV [33] in a venous vascular territory. It should be noted that venous thrombosis can proceed to venous infarction; in that circumstance, the perfusion CT findings are the same as arterial infarction.

Recently, a number of incidents have been reported in which patients were mistakenly administered up to eight times the normal radiation dose during perfusion CT imaging performed to diagnose stroke. These instances were due to incorrect settings on the scanner console. About 40% of the patients lost patches of hair as a result of the radiation overexposure. These episodes emphasize the importance of CT quality assurance programs. Radiologists and technologists should be familiar with the doses normally displayed on the scanner console. Radiologists and physicists should ensure that all CT protocols respect the as low as reasonably achievable dose principle [34].

Perfusion CT studies are optimally performed at 80 kVp and 100 mAs [35], for which the effective radiation dose associated with a single-slab perfusion CT study is approximately equal to that of an unenhanced head CT (approximately 2–3 mSv). These parameters provide the advantage of not only reduced radiation dose but also improving image quality. For instance, perfusion CT using 80 kVp rather than 120–140 kVp not only reduces the administered radiation dose by a factor of ∼ 3 but also increases conspicuity of IV contrast material. The latter finding is due, in part, to the fact that the photoelectric effect for 80-kVp photons is much closer to the K-edge of the iodine within the contrast material [35].

A comprehensive stroke CT protocol that includes unenhanced and contrast-enhanced head CT, perfusion CT, and CTA of the cervical and intracranial arteries can deliver a large mean effective dose (on the order of 7–8 mSv) [36]. However, not every patient requires all the components of such a protocol. Instead, dedicated stroke protocols should be tailored to specific clinical indications, especially in stroke patients who may need multiple imaging studies involving ionizing radiation (e.g., unenhanced head CT, CTA, perfusion CT, and digital subtraction angiography) during a single admission. In such patients, radiation reduction strategies should be implemented as appropriate. MRI, when feasible, should be used in the nonemergent subsequent imaging evaluation of stroke patients.

As in all of medicine, the potential risks of any diagnostic test or therapeutic procedure (however rare) must be weighed against the very real benefits of limiting disability and preventing death. Perfusion CT imaging, when appropriately and correctly performed, is justified and provides safe valuable information that can substantially contribute to the management of acutely ill patients with acute cerebrovascular disease.

Multimodal CT that includes perfusion CT is playing an increasingly important role in the initial evaluation of acute ischemic stroke patients. The perfusion CT technique is safe when performed according to a standard acquisition protocol that uses 80 kVp and no more than 100 mAs. It is important for radiologists using perfusion CT for stroke imaging to be familiar with the postprocessing software used at their institution, including its advantages and limitations. Radiologists should be familiar with the parameters that can be selected during the postprocessing and how these may influence the perfusion CT results. When reading a perfusion CT study, radiologists should keep in mind that not every perfusion abnormality seen on perfusion CT represents an ischemic stroke and be aware of the most common stroke mimics.

Address correspondence to M. Wintermark ([email protected]).