Applied Radiology

Dr. Zaharchuk is a fourth-year Radiology Resident at the University of California at San Francisco, CA. He received his MD from Harvard Medical School, Boston, MA, in 2000 and his PhD in Applied Physics from Harvard University in 1999. He will begin a Fellowship in Neuroradiology in 2005.

Alterations in cerebral blood flow (CBF) lie at the core of many brain disorders, most notably stroke. Imaging the regional distribution of CBF quantitatively remains challenging. ^1,2 Fundamental differences in accuracy and artifacts exist between techniques that employ intravascularly confined versus freely diffusible tracers. ³ Currently, H ₂ ¹⁵ O positron emission tomography (PET), which uses a freely diffusible tracer, is the perfusion gold standard, given the ability to quantitate CBF with relative insensitivity to vascular transit time variations. ⁴ However, this examination is costly, entails radioactive dosing and invasive monitoring, and has low intrinsic spatial resolution. In addition, relatively few H ₂ ¹⁵ O PET imaging centers exist, due to the difficulties of staffing and maintaining an onsite cyclotron. Thus, there is great interest in adapting more widespread imaging modalities, such as magnetic resonance imaging (MRI), to measure quantitative CBF.

The most widespread, clinically applicable MRI technique for estimating cerebral hemodynamic parameters is dynamic susceptibility contrast, which has also been termed "perfusion-weighted imaging." In this technique, a rapidly administered dose of an intravascular contrast agent, such as gadolinium diethylenetriamine penta-acetic acid (Gd-DTPA), is injected using a power injector into a peripheral vein at 5 mL/sec, and the passage of this bolus through the brain is monitored using rapid T2*-sensitive MR imaging. This is quite efficacious for measuring relative cerebral blood volume (CBV) but requires delicate deconvolution methods to extract CBF. ⁵ New approaches to perform this deconvolution, particularly in the presence of delay and dispersion of the arterial profile of the bolus, may increase the accuracy of measurements in pathological brain tissue. Also, steady-state susceptibility contrast using intravascular tracers with long blood half-life ("blood pool" agents) offers the potential of higher signal-to-noise ratio (SNR) and increased fidelity of CBV measurement in the setting of a disrupted blood-brain barrier (BBB). ^6,7

More recently, arterial spin labeling (ASL) methods to measure quantitative CBF without the use of intravenous contrast agents have been pro-posed. ⁸ Proton-density brain images are acquired with and without labeling of the inflowing blood; direct subtraction of these images is roughly proportional to CBF. Pitfalls regarding quantitation revolve around regional differences in the mean arrival time of the arterial blood to the tissue voxel. ⁹ Because the label is a magnetic one, it has a half-life of blood T1 (approximately 1.2 to 1.4 seconds at 1.5T), which can have a significant effect on the measured CBF. Recent advances include approaches to measure CBF and arterial arrival time simultaneously and new methods of performing the labeling, such as velocity-selective ASL, ¹⁰ which may reduce the importance of transit delays. Regional perfusion territory imaging is an ASL technique in which brain-feeding arteries are selectively labeled, permitting visualization of individual arterial territories, ¹¹ which can vary from person to person. ¹²

Several new techniques have shown high potential for cerebral perfusion imaging studies. The stable isotope oxygen-17 has a nuclear spin and can interact with protons on water molecules. It can be used to label water (H ₂ ¹⁷ O), allowing CBF measurement using washout kinetics, such that differences in arrival times are largely irrelevant. Finally, new approaches using hyperpolarized water-soluble gases, such as ¹²⁹ Xe, or hyperpolarized ¹³ C on organic molecules, offer whole new paradigms for perfusion MRI. ^13,14 Table 1 summarizes the relative advantages and disadvantages of these myriad techniques.

Susceptibility contrast, dynamic ("bolus") and steady-state

Belliveau and coworkers ¹⁵ described the first approach to the use of lanthanide chelates as intravascular MRI tracers to measure relative CBV. These agents are retained within the vasculature of the brain by the intact BBB, and during their passage through the brain vasculature, they cause an increase in transverse relaxivity that can be detected by using rapidly acquired T2*- or T2-weighted images (on the order of 1 to 2 images per second) (Figure 1). The major mechanism of contrast was proposed to be the dephasing of water protons diffusing within magnetic field gradients created by the difference in magnetic susceptibility between the contrast-filled vessel and the rest of the tissue. ¹⁶ Based on modeling studies, Fisel and colleagues ¹⁷ predicted that T2*-weighted images acquired by using gradient-echo images had equal sensitivity to vessels of all sizes, thus allowing total CBV to be measured. Alternatively, due to the rephasing of the spin echo, T2-weighted images were more sensitive to water diffusing around vessels on the 5- to 100-µm size, affording sensitivity to changes in micro-vascular or capillary CBV. Dynamic susceptibility contrast perfusion imaging has become a widespread tool in clinical MRI studies, due to the relative simplicity of the method and the fact that postcontrast T1weighted images are frequently obtained as part of many standard MRI brain imaging protocols.

Cerebral blood volume

Cerebral blood volume, defined as the fraction of the imaged voxel comprising the intravascular space, ranges from 2% to 5% in humans. ¹⁸ It is a potentially sensitive indicator of the vascular endothelial response to changes in the local CBF and tissue metabolism. Tracer kinetic principles state that relative CBV can be measured by the area under the curve (AUC) of the voxel concentration versus time, while absolute CBV can be determined by dividing the voxel AUC by a "reference" voxel known to contain 100% blood, such as the superior sagittal sinus. ¹⁹

Several technical issues can impact the accuracy of this relatively simple measurement. The first is that dynamic susceptibility contrast arises from spin diffusion in the space surrounding the blood vessels, such that it is difficult to determine a reference voxel for a 100% blood-filled voxel. Next, the effects of contrast recirculation must be eliminated or minimized. Fitting of the concentration time curve to a suitable function (such as a gamma variate c(t)=Kt ² e ^-t ) has been frequently applied, though it entails increased computational time and decreases the SNR of the measure-ment. ²⁰ Also, CBV measurements can be erroneous if Gd-DTPA leaks through a disrupted BBB. ²¹ This is of particular concern in tumor research, because many tumors cause BBB disruption, causing a bright signal on postcontrast T1 images.

Imaging with traditional gadolinium (Gd)-based compounds after the blood contrast is more uniform can avoid some of these issues, but this is a relatively low SNR technique. ²² New molecules, such as superparamagnetic iron-oxide particles, dendritic compounds saturated with Gd atoms, or reversible protein-binding Gd-based agents have a longer half-life in the blood. ^23-25 These so-called "blood pool" agents offer significant improvements in steady-state imaging, which has found initial application in T1-based angiographic sequences. However, they also offer a higher SNR method for measuring CBV using steady-state susceptibility contrast that circumvents many of the issues outlined earlier. Favorable results in several late-stage clinical trials in both the United States and Europe were reported at the International Society for Magnetic Resonance in Medicine (ISMRM) 12th Scientific Meeting and Exhibition meeting in Kyoto, Japan. ⁷

Cerebral blood flow

Cerebral blood flow is the primary rate constant controlling the supply of nutrients and the removal of waste products from the brain. Below a threshold level, the product of the time duration and absolute CBF level can predict tissue infarction. ²⁶ While the CBV measurement with intravascular tracers is relatively straightforward, measurement of CBF is more challenging. ²⁷ Typically, one tries to deconvolve the effects of a bolus of finite width (as estimated from the arterial signal near a large feeding vessel, such as the anterior communicating artery). This arterial input function is assumed to represent the profile of the bolus at its point of entry into each individual voxel. The assumption is made that every voxel has the same arterial input function. This is clearly an oversimplification, and uncertainties about the regional changes in the profile and timing of the bolus can cause significant errors in CBF. ²⁸ Despite this, good absolute CBF correlation with H ₂ ¹⁵ O PET was shown in anesthetized, healthy pigs, ²⁹ and excellent relative CBF correlation was reported in an experimental ischemia model. ³⁰

In the diseased brain, where regional arterial stenoses, occlusions, and collateral pathways are more common, intra-vascular tracer CBF measurements that do not account for variations in bolus delay and dispersion are inaccurate. ^27,31 Recent studies have shown that simple curve shifting to simulate delays leads, in general, to CBF underestimation, but with an oscillatory component, making back calculation of CBF extremely dif-ficult. ²⁸ In humans with unilateral carotid occlusion, the relative CBF correlation with H ₂ ¹⁵ O PET imaging was good, but the proportionality constant between the 2 techniques varied from subject to subject, casting doubt on the accuracy of absolute CBF measurement with dynamic susceptibility contrast. ³¹ These errors are clinically relevant, as evidenced by changes in the sensitivity of specificity of CBF measurements using dynamic susceptibility contrast in human stroke based on whether the ipsilateral or contralateral arterial input function was selected. ³²

As alluded to earlier, in theory at least, the effects of delay can be accounted for by simple time shifting of the voxel curves, such that the arterial input function and the voxel contrast versus time curves begin at the same time. However, because these curves are typically sampled on the order of 1 to 2 seconds, differences on this order cannot be further corrected. Also, it is often difficult to estimate the precise arrival time of the bolus in an area with low flow, because SNR is poor in these regions. Addressing the effects of dispersion is even more challenging, because they cannot be predicted based on changes in arrival delay. At the 2002 ISMRM meeting, Alsop et al ³³ suggested that it might be possible to find vox-els throughout the image that could act as "local" arterial input functions. Rather than defining a single ROI as the arterial input function for the entire brain, an attempt is made to identify "arterial-like" voxels located either spatially closer to the imaged voxel or at least within the expected vascular territory. Such an approach may more closely realize the true profile of the contrast bolus as it enters the voxel, accounting for both delay and dispersion. Various schema have been suggested to select these local arterial input functions, including early arrival, high contrast first moment, narrow full-width half-maximum, high peak concentration, and relatively high blood volume. Alsop et al ³³ defined a cube surrounding each voxel from which the most likely local arterial input function was chosen; this is then repeated for each voxel within the brain. Their results suggested that differences in local arterial input function could be discerned in the setting of ischemic stroke.

At the 2004 ISMRM meeting, Lorenz and colleagues ³⁴ confirmed that such a local arterial input function method led to increased CBF levels in those voxels that would be assigned an extremely low CBF if a global arterial input function method were used, as expected if the effects of dispersion and delay were reduced. Ostergaard and colleagues have suggested a slightly modified local arterial input function method that searches the entire brain for arterial-like voxels (personal communication, June, 2004). A standard vascular territory atlas is then used to match the imaged voxel with the appropriate arterial input function. This method has the advantage that the selected arterial voxels are less likely to be contaminated by partial voluming with "tissue" contrast curves, but assumes standard vascular territory anatomy, which may not be true in the setting of vascular occlusions or variations in the circle of Willis. Such local arterial input function methods offer hope that a better characterization of the bolus as it enters the voxel may be possible, and with it, that dynamic susceptibility contrast MRI methods for measuring CBF in the setting of abnormal vasculature may be made more dependable.

Arterial spin labeling

Arterial spin labeling is a noncontrast method in which protons attached to water molecules in blood are magnetically labeled before entering the brain tissue; after labeling, these protons distribute in proportion to the local CBF to each voxel (Figure 2). ⁸ The use of water (which is freely diffusable and extracted at >90% during its first arterial passage at normal CBF) as the label confers important advantages over the use of intravascular tracers, including the potential to be quantitative, repeatable, and independent of the status of the BBB. ¹ The ease of repeatability of the method lends itself to use in CBF challenge studies (Figure 3). Difficulties associated with CBF quantitation are related to the relatively short "halflife" of the tracer (blood and tissue T1) and uncertainties regarding the precise timing of the arrival of the labeled water at different voxels. ⁹ In normal subjects, ASL has been favorably compared with H ₂ ¹⁵ O PET. ³⁵

All current ASL methods are based on the collection of image pairs, subsequently subtracted, in which arterial magnetization entering the imaging slice during the repetition time (TR) interval varies. For maximum contrast, inflowing spins should have equilibrium magnetization (+M ₀ ) during the control image and be inverted (-M ₀ ) during the label image. In practice, the difference signal between the label and control images is a tiny fraction (0.5% to 2%) of the source images, requiring that multiple image pairs be collected and their small signals averaged to provide adequate sensitivity. Labeling can be achieved using either a short radiofrequency pulse to invert the spins followed by a delay time to allow inflow ("pulsed ASL" [PASL]), ^36-38 or by continuous adiabatic inversion of spins crossing a predefined plane, defined by off-resonance low-level continuous wave radiofrequency radiation in the presence of a magnetic field gradient ("continuous ASL" [CASL]). ⁸ In theory, CASL permits an approximately 3fold increase in SNR, though this advantage is partially mitigated by imperfect adiabatic labeling and longer distances between the labeling plane and the imaged slices. ³⁹ Because the magnetic label decays with the blood T1 (1.2 to 1.4 seconds at 1.5T), both methods suffer from CBF underestimation in regions with prolonged arterial arrival times, defined as the average time needed for the labeled blood to reach the capillary level, measured to be between 400 and 1500 msec in the normal brain. ⁴⁰ This artifact can be minimized by inserting a postlabeling delay before imaging in CASL or with saturation pulses in PASL. ^9,41 However, in the diseased brain, arterial arrival times may be markedly increased (up to 5 seconds more) if flow is provided by collateral networks; in this case, by the time the label arrives, its has relaxed back to equilibrium, and an erroneously low CBF will be measured. If diffusion gradients are not used to suppress signal from large feeding arteries, regions with a delayed arrival time often show multiple, serpiginous bright spots, with decreased flow signal distal to the supplying vessels (Figure 4). All of these methods to improve quantitation lead to a loss in the SNR of the acquired perfusion images.

Surmounting regional arrival time uncertainties

At the 2004 ISMRM meeting, 2 approaches to more accurately measure CBF in the presence of unknown regional changes in the arterial arrival time were described. It is possible to image the water inflow at multiple postlabel delay times, which allows visualization of the movement of labeled blood from the arteries to the parenchyma, but this is generally not time efficient (Figure 4). However, if the full wash-in curve is measured, arrival time and CBF can be independently measured. ⁴² At the most recent ISMRM meeting, Guenther and colleagues ⁴³ showed a time-efficient means of measuring this wash-in, using a 3-dimensional fast spin-echo approach. Images from 5 different delay times could be acquired during a single 5minute acquisition. Movies were created showing the passage of the labeled blood, essentially quantitative, tomographic versions of conventional cerebral angiography. For each voxel, curve fitting for both CBF and arterial arrival time can be performed. This permits theoretically improved CBF measurement, but also produces a map of arrival time differences, which yields similar information to bolus "time-to-peak" maps, which may also be clinically meaningful.

A second approach to the issues surrounding heterogeneous arterial arrival time delays has been suggested by Wong and colleagues, ¹⁰ in which a 90 -gradient-180 _y -gradient-90 _x labeling pulse can be used to create a label independent of position but sensitive only to spin velocity, which they have termed velocity selective ASL (VSASL). By labeling moving spins even within the imaged voxel, arterial arrival time effects are likely greatly reduced. During the composite labeling pulse, the gradient is alternated between a high and low level, imparting differential labeling of spins moving faster than a "critical" velocity, v _c = π*γδ∆ G , where γ is the gyromagnetic constant for protons (42.6 MHz/T), δ is the length of the gradient pulses, ∆ their separation, and G is the gradient strength. ⁴⁴ After labeling, a time τ is allowed for the movement of this label into the tissue, and images acquired with a diffusion gradient permit imaging of only those spins that have been extracted into the tissue (Figure 5). Initial implementations measured relatively low CBF levels (28 mL/100 g/min) in normal brain and were confounded by slow-flow CSF artifacts in regions surrounding the ventricular system. ⁴⁴

At the ISMRM meeting, several refinements were demonstrated. The first was a time-efficient method for CSF nulling using a VSASL approach, ⁴⁵ which was shown to reduce artifacts seen around the ventricles. The other was an examination of the directional dependence of the VSASL approach, because spins must have velocity components in the direction of the gradient for effective labeling. ⁴⁶ It was reported that direction effects were significant in the range of v _c from 1 to 3 cm/sec, and that taking this into account may permit improved quantitation. Velocity selective ASL approaches have great promise to minimize or eliminate issues surrounding arrival time differences, which plague quantitative ASL approaches, and may be extremely useful for imaging in the setting of carotid occlusion or stroke.

High-field ASL

Arterial spin labeling is an ideal application for high magnetic-field imaging machines. This is due to the intrinsic increase in SNR of such machines, the use of proton-density-weighted images, and the longer T1 of blood, all of which contribute to increased sensitivity to the perfusion label with less sensitivity to uncertainties in the arterial arrival time. For example, at 3T, there is roughly 2-fold higher intrinsic SNR. The blood T1, which determines the half-life of the label, increases significantly from about 1.2 to 1.4 seconds to 1.6 to 1.8 seconds. This leads to a boost in SNR (because the label decays more slowly), but, perhaps more importantly, to decreased sensitivity to arterial arrival time differences by enabling the use of longer postlabeling delays. Altogether, the increased SNR at 3T corresponds to a 4- to 6-fold decrease in the required imaging time to acquire similar sensitivity (Figure 6). Several reports at the ISMRM meeting specifically addressed and vividly confirmed the improvement in continuous ASL perfusion imaging at 3T.

Talagala and coworkers ⁴⁷ showed the high SNR of a 3T system using continuous ASL with a 16-element phased-array receive coil and separate neck coils for labeling. They found that images requiring only 1 to 2 minutes to acquire had adequate SNR for 3-mm isotropic resolution. Importantly, their multicoil approach is well suited for application to even higher magnetic-field scanners. Wang and co-workers ⁴⁸ also reported improved-quality ASL perfusion images on a 3T system, using a single coil, amplitude-modulated control pulse to control for the effects of magnetization transfer. They showed that it was possible to stay within specific absorption rate (SAR) guidelines for patient heating while maintaining reasonable labeling efficiency. They showed a 33% improvement in SNR for this technique compared with a pulsed ASL method at the same field strength.

Regional perfusion territory imaging

This relatively new ASL-based imaging modality is based on the concept of spatially selective labeling, which can produce images of the perfusion territory supplied by the labeled vessels. Such images permit individual imaging of vascular territories, which have been shown to be variable from person to person. ¹² Applications may include analysis of patients with high-grade or complete large vessel vascular occlusions to better inform the use of angioplasty or stenting, evaluation of cerebrovascular bypass grafts, or estimation of the likelihood of an embolic or atherothrombotic source in a patient with multiple bright regions on diffusion-weighted imaging.

My colleagues and I ¹¹ first reported on this method in humans by using continuous selective labeling at the level of the carotid bifurcation using a small, separate labeling coil. This separation of the labeling and imaging part of the ASL experiment also allowed for lower radiofrequency power deposition and independence from magnetization transfer issues, thus simplifying quantitation. This initial description required a "home-built" second transmitter channel, and has yet to be incorporated into a standard clinical system. Because of this, approaches to selectively label arteries by using standard product head coils have been more recently developed, several of which were described at the most recent ISMRM meeting.

The first uses a magnetic-field gradient applied in 2 simultaneous directions during a continuous ASL labeling period, which creates a tilted labeling plane. ⁴⁹ The gradient in the coronal direction was manipulated such that the resonance frequency of the plane was at the normal location on the labeling side, and canted downward on the nonlabeled side, such that the on-resonance position was effectively outside the radiofrequency volume of the head coil. Because there is no radiofrequency power to adiabatically invert flowing spins in this location, no labeling occurs. This method retains the desirable characteristics of continuous ASL, but is highly dependent on geometry, and current methodology allows only labeling of the anterior circulation.

Two abstracts describing pulsed ASL methods were shown. Song and colleagues ⁵⁰ created a label pulse encompassing either the left or right hemisphere, and demonstrated that perfusion territory maps of the anterior circulation could be created. Hendrikse and coworkers ⁵¹ showcased an elegant method in which a preselected rectangular volume can be selected for labeling, based on a previously obtained MR angiogram. Excellent separation of the left carotid, right carotid, and posterior circulation was shown (Figure 7). Variations in normal subjects based on differences in the circle of Willis were noted. In addition, the method has been applied to study the perfusion territories of extracranial-intracranial (EC-IC) and superficial temporal artery-middle cerebral artery (STA-MCA) bypass grafts.

This information has been previously available only in a qualitative way using invasive cerebral catheter angiography. It is likely that such methods may yield more complete understanding of the cerebrovascular dynamics in patients at high risk of stroke or being considered for corrective surgical or neurointerventional procedures.

Oxygen-17 water

Oxygen-17 is a nonradioactive isotope that occurs naturally at low concentration (0.038%). When enriched and incorporated into a water molecule (H ₂ ¹⁷ O), it can be used as a weak MRI contrast agent, either imaged directly with a dedicated ¹⁷ O receiver coil ⁵² or through the decrease in T2 it causes in protons (Figure 8). ^53,54 Such a contrast agent has a high safety profile, since H ₂ ¹⁷ O has identical chemical properties of normal water. This molecule also has ideal physiochemical characteristics to image CBF, being freely diffusible into the extravascular space, stable, and nonradioactive. The mathematical analysis developed for H ₂ ¹⁵ O PET imaging can be directly applied to the ¹⁷ O method-most importantly, the ability to use a simple washout measurement of CBF. Such a measurement would be largely independent of delay and dispersion, typically measured in seconds, because the washout time is on the order of minutes for typical CBF levels in humans. Also, advantages would accrue from the larger number of centers that can perform MRI than hemodynamic PET, and correlation with other MRI sequences acquired during the same session would be straightforward.

The largest barrier to the use of H ₂ ¹⁷ O MRI is the current high cost of enrichment, currently on the order of $2000 per 100% enriched g H ₂ ¹⁷ O. It is likely that this price would decrease given increased demand. Because of this, most studies documenting its efficacy for CBF measurement have required intra-arterial injection. Initial studies in small animals have documented feasibility and sensitivity to changes in carbon dioxide partial pressure, known to be a potent CBF enhancer, and in the setting of focal ischemia. ^55,56 More recently, this technique has been applied to larger animals, such as the macaque monkey. ⁵⁴ At the 2004 ISMRM meeting, CBF was measured in beagles with H ₂ ¹⁷ O MRI using a protocol identical to the H ₂ ¹⁵ O PET experiment, including intravenous rather than intra-arterial injection; this study showed excellent correlation with H ₂ ¹⁵ O PET. ⁵⁷ It is likely that H ₂ ¹⁷ O will soon find application for measuring CBF in patients undergoing cerebral angiography and neurointerventional procedures in centers with joint MRI-angiography suites.

Hyperpolarized approaches

Conventional proton MRI depends on the high concentration of water protons (80 M), because even at high clinical magnetic fields, the relative polarization (ie, the number of excess spins directed along the main magnetic field) is very small, on the order of 1 in every million spins.

Various methods exist to create a nonequilibrium state in which nuclear polarization is increased far beyond this small baseline-as high as 50%- such that imaging of nuclei with far lower concentration than protons becomes feasible. At the ISMRM meeting, impressive advances in this arena were shown.

In the mid-1990s, it was shown that hyperpolarization of noble gases (such as ³ He and ¹²⁹ Xe) could be achieved by optical laser pumping in the vapor form, with polarizations as high as 10% to 40%, establishing the field of hyperpolarized gas MRI. ⁵⁸ While initial applications targeted lung ventilation, breathing of ¹²⁹ Xe or encapsulation within a lipid emulsion has been proposed as a means of measuring CBF. ^13,59,60 This nucleus has several advantageous features, including a long in vivo T1 (20 to 30 seconds), which allows transport from the lungs or intravenous system to the brain without undue loss of polarization. Also, ¹²⁹ Xe has a strong chemical shift based on its local environment, and the chemical shift offset between xenon in arterial blood-borne liposomes versus that extracted into the brain can be resolved, theoretically permitting truly local arterial input function measurement, leading to improved quantitation in perfusion imaging. At the ISMRM meeting, ¹²⁹ Xe administered by face mask was used to detect CBF changes due to hypercapnia. ⁶¹

Several promising hyperpolarization approaches based on electron-proton spin exchange have been recently described. These rely on the concept that the manipulation of electron spins using microwave energy can be transferred to the nuclear spins, thus creating large nonequilibrium nuclear magnetization. Typically, the use of microwave radiofrequency excitation has limited this application in humans, due to undesirable heating. Recently, a system has been shown in which high degrees of polarization (up to 37%) of ¹³ C and ¹⁵ N nuclei can be created at an extremely low temperature in the solid state, and then quickly dissolved into the liquid state at body temperature, with retention of their polarized state. ¹⁴ Such a method offers increases in SNR for these nuclei on the order of 10,000, with a sensitivity approaching or even exceeding conventional proton MRI. Angiographic and cardiac perfusion images in the rat using ¹³ C-labeled urea with higher SNR than conventional proton or Gd-enhanced MR angiography have been shown. ⁶² At an ISMRM plenary session, further impressive angiographic and perfusion images were shown in a pig model (Figure 9). ⁶³ Recently, this approach has been applied to measure brain perfusion by using bolus tracking methods with ¹³ C attached to an intravascularly confined small molecule. ⁶⁴ One advantage of these techniques is that the radiofrequency coils are tuned to the ¹³ C resonance frequency, so that there is no background signal from the static tissue. This permitted the use of direct 2D projections, which may be acquired more rapidly than 3D or stacks of 2D images. In addition, because the hyperpolarization is destroyed at the end of the pulse sequence, contamination due to recirculation is avoided. Finally, because the high polarization is independent of imaging field strength, it is conceivable that smaller magnets or even the earth's magnetic field could be used for imaging, enabling a whole new class of portable MRI systems, as might find application in a neurological intensive care unit.

Conclusion

Advancements in cerebral perfusion MRI are being realized on many fronts. At the most clinical level, increasingly sophisticated modeling of intravascular Gd dynamic and steady-state susceptibility contrast offers the possibility of more accurate CBF and CBV methods in patients with cerebrovascular disease. Both fundamentally new methods of labeling and signal acquisition as well as the use of high-field imaging systems are leading to improvements in ASL-based CBF measurements. Regional perfusion territory imaging of individual arterial vascular beds is a new modality that may yield fresh insights into the disrupted cerebrovasculature. H ₂ ¹⁷ O imaging offers the potential of an MR equivalent to the more expensive and cumbersome H ₂ ¹⁵ O PET methodology. Finally, unexpected advances using dynamically hyperpolarized nuclei such as ¹³ C and ¹²⁹ Xe may lead to increased SNR and possibly even portable MRI systems. Such a plethora of MR approaches to the measurement of cerebral perfusion is a testament to the clinical importance of the measurement, the flexibility of the MR experiment, and the genius and hard work of the worldwide MR research community.