Physics

The imaging technique most often used for DWI is an echo planar imaging (EPI) sequence. In EPI, two separate and equal magnetic gradients are applied to opposite sides of the radiofrequency (RF) pulse during image acquisition. Use of these bipolar pulsed gradients allows detection of diffusional motion by changes in the magnitude of the moving spins from phase dispersion. Water molecules that do not travel significantly between excitation pulse and read pulse will not dephase significantly and will therefore retain much of their initial signal. Thus, in pathologic states, the increased signal on DWI reflects an abnormal decrease in water diffusivity, typically caused by either loss of normal extracellular space (as seen in cytotoxic edema) or the presence of a highly viscous, proteinaceous, or cellular environment.^1,2 It is worth noting that loss of signal is not due to travel of water from one imaging voxel to another, but rather motion within the voxels themselves. Indeed, the average motion of a water molecule at 40°C is 2.5 × 10⁻³ mm²/sec, which translates into motion of 22 µm in 100 msec; this is considerably smaller than the typical 1- to 2-mm voxel dimension.³

The magnitude of diffusion sensitization of DWI is determined by the “b-value,” which in turn is related to the duration, strength, and time interval between the magnetic gradients. Therefore, higher b-values confer more sensitive DWI and thus yield improved contrast and ability to identify areas of water restriction; however, higher b-values also result in loss of signal, as well as noisier images, which may then have reduced utility. The b-value that is typically used in clinical assessment is approximately 800 to 1000 sec/mm²,⁴ but with certain applications, such as vertebral imaging or characterization of intracranial tumors, b-values may range from 500 to 2000 sec/mm² or greater.

The rate of diffusional motion is characterized by the apparent diffusion coefficient (ADC). This quantitative estimate of diffusivity can be achieved by acquiring two sets of images with different b-values, which can eliminate the effects of spin density and T1 and T2 relaxation. In clinical practice, one of the b-values would measure approximately 1000 sec/mm², as indicated earlier. The other b-value is typically 0 sec/mm², which reflects an image that does not have any motion-probing gradient applied (yielding so-called B0 images). ADC maps quantify differences between B0 and B1000 images and therefore nullify “T2 shine-through,” a condition in which high B1000 signal is primarily due to high T2 signal rather than water restriction (Fig. 19-1). Thus, true restricted diffusion, as seen with acute ischemia, will demonstrate high signal on DWI and corresponding low signal on ADC maps.¹

FIGURE 19-1 Breast carcinoma metastasis producing a T2 shine-through artifact. A, Axial diffusion trace image demonstrating a rounded focus of increased diffusion signal in the right posterior frontal lobe (asterisk). B, Corresponding T2 image (B0 image) showing increased signal in the same distribution (asterisk), thus suggesting that the diffusion signal may not reflect true restriction. C, Apparent diffusion coefficient map confirming that this area is not dark, so the lesion does not truly restrict diffusion.

Clinical Uses and Applications

In clinical use, diffusion images are acquired in at least three directions, with the resulting images mathematically averaged to generate the trace image in which regions of normal white matter and gray matter appear fairly uniform. The normal brain shows slight variation in diffusion signal based on cellular structure. Gray matter has marginally higher signal on DWI, probably reflecting differing T2 properties rather than true differences in the ADC. Areas that are essentially water (most notably CSF) are dark on DWI sequences because of the free motion of water molecules in bulk water.⁵

The most completely characterized application of DWI is for the diagnosis and management of patients with acute cerebral ischemia. When perfusion does not meet the metabolic demand of territorial parenchyma, energy-dependent sodium-potassium adenosine triphosphatase ion pumps may fail. The resulting ion flux leads to the accumulation of water within the intracellular compartment (cytotoxic edema). Neuronal swelling crowds the extracellular space, with resulting restricted motion of extracellular water molecules that is manifest as high DWI signal; these changes may be seen within minutes of tissue infarction. Normal brain ADC values range from approximately 740 to 840 × 10⁻⁶ mm²/sec, with considerable overlap between white matter and gray matter values.⁶ Parenchyma with ADC values of less than 500 to 550 × 10⁻⁶ mm²/sec is almost certainly in the process of irreversible infarction. Intermediate ADC values (550 to 700 × 10⁻⁶ mm²/sec) may reflect parenchyma that is ischemic yet still viable; it is this tissue that may be potentially salvageable with appropriate intervention. Although DWI is frequently identified as the most sensitive sequence for detecting infarction, other complementary sequences (such as the perfusion sequences described later) can help define an apparent impending infarction and define tissue at risk for infarction, the so-called ischemic penumbra.^7,8

As infarcts age, the ADC and DWI patterns mature at differing rates. Although both become abnormal within minutes, the ADC increases from low signal to isointensity more rapidly, in most instances within 5 to 14 days. The ADC then continues to rise over the ensuing weeks to remain positive for the life of the patient, corresponding with encephalomalacia. DWI normalizes more slowly. It may reach isointensity with surrounding brain as long as 4 to 5 weeks after the infarction, and it will take even longer for signal to decline further in the setting of encephalomalacia (Fig. 19-2).

FIGURE 19-2 Acute left lenticular infarct. A, Axial non–contrast-enhanced computed tomography (CT) showing a focal chronic lacunar infarct (arrowhead), but no area suspicious for acute infarction. B, Axial T1-weighted image obtained 3 hours after CT showing no definite infarction. C, T2-weighted image showing an area of vague lenticular edema (arrow) distinct from the chronic lacunar infarct (arrowhead). D, A fluid-attenuated inversion recovery (FLAIR) T2-weighted image also shows the lenticular edema (arrow). The chronic lacunar infarct (arrowhead) is less conspicuous on FLAIR than on T2 imaging, probably related to the canceling effect of volume loss and gliosis on this sequence. E, Diffusion-weighted imaging showing increased signal corresponding with the infarct (arrow) in C and D, whereas the chronic focus (arrowhead) has no restriction. F, The apparent diffusion coefficient map is dark in the area of edema, thus confirming true restriction of diffusion (arrow). The lacunar infarction is mostly bright, which suggests facilitated diffusion, as is seen with volume loss (arrowhead).

Certain intracranial neoplasms demonstrate restricted diffusion. Many of these neoplasms are tumors that are high in cellularity, with resulting high signal on DWI and low signal on the ADC map (i.e., true diffusion restriction). Examples include lymphoma, medulloblastoma, and portions of high-grade gliomas. Some tumors demonstrate restriction without high cellularity. Hence, although both epidermoid cysts and arachnoid cysts closely follow fluid intensity on T1, T2, and fluid-attenuated inversion recovery (FLAIR) sequences, the high DWI signal and heterogeneous low signal on the ADC map conferred by desquamated debris allow confident diagnosis of an epidermoid. Arachnoid cysts, which instead contain simple fluid, demonstrate facilitated diffusion and therefore have the opposite diffusion pattern (low DWI signal and high ADC signal).⁹

Brain abscesses are typically peripherally enhancing lesions that are usually surrounded by significant vasogenic edema. Occasionally, these lesions can be difficult to differentiate from other peripherally enhancing masses such as necrotic tumors. The central cavity of an abscess demonstrates high signal on DWI and low signal on ADC maps. The restricted diffusion is most likely attributed to the high viscosity of proteinaceous fluid and the hypercellularity of inflammatory cells.^10–12

Seizure activity can induce changes in water diffusivity because of cellular swelling and fluctuation in extracellular fluid. Cortical restricted diffusion has been seen with prolonged seizure activity, presumably caused by an imbalance between oxygen delivery and consumption.¹³ Focal parenchymal changes in the postictal state can be seen as hyperintense signal changes on T2-weighted sequences with variability in DWI signal. In the setting of ischemia related to seizure activity, DWI may initially demonstrate restricted diffusion as a result of cytotoxic edema, but the restricted diffusion can later progress to decreased diffusion signal, which may represent gliosis of the involved tissue.¹⁴

DWI also aids in the diagnosis and further characterization of other disease processes. Restricted diffusion can be seen in zones of active demyelination in such disease processes as multiple sclerosis and progressive multifocal leukoencephalopathy (Fig. 19-3), in areas of active inflammation typically seen in the medial temporal lobes in herpes simplex encephalitis, in the anterior deep gray nuclei of the basal ganglia and along a discontinuous cortical ribbon in Creutzfeldt-Jacob disease, in some phases of hemorrhage, and in some encephalopathies and leukodystrophies.^1,9

FIGURE 19-3 Forty-two-year-old man infected with human immunodeficiency virus and progressive multifocal leukoencephalopathy. A, Axial fluid-attenuated inversion recovery T2-weighted image showing a nonspecific area of high signal, possibly edema, near the vertex (arrow). B, Diffusion-weighted imaging showing a small area of corresponding signal abnormality, with restricted diffusion noted preferentially at the periphery (arrow). C, A slight increase in signal on the apparent diffusion coefficient map suggests T2 shine-through (arrow). Ordinarily, progressive multifocal leukoencephalopathy does not enhance, but with the polydrug regimen and immune reconstitution, one may see more distinctive restriction of diffusion and enhancement, both of which are absent in more indolent cases.

As indicated earlier, facilitated diffusion can be seen in patients with encephalomalacia and arachnoid cysts, with resulting low signal on DWI and high signal on ADC maps. Other entities with a similar pattern include cystic lesions, vasogenic edema, transependymal spread of CSF in hydrocephalus, tumor necrosis, and radiation necrosis.

Pitfalls and Limitations

EPI techniques allow ultrafast acquisition times, thereby almost eliminating motion artifact. However, the long echo train lengths needed to obtain the data render these sequences sensitive to both chemical shifts and magnetic susceptibility. Lipid suppression can be used to resolve some effects of chemical shift. Eddy currents result from the use of rapidly alternating gradients; these currents may result in significant distortion or misregistration between directional acquisition, which translates into blurring and loss of soft tissue contrast in the resulting trace image. Traditional DWI is prone to areas of susceptibility, most notably about the skull base, paranasal sinuses, and petrous bone, where air-bone-tissue interfaces are present.^5,15,16 In addition, the presence of paramagnetic and ferromagnetic material such as blood products or metal from surgery, trauma, or dental hardware can produce significant artifact, seen as ghosting, image distortion, and susceptibility artifact (Fig. 19-4). Newer techniques (periodically rotated overlapping parrallel lines with enhanced reconstruction [PROPELLER], BLADE, and others) use a rotating acquisition frame to minimize the effect of such artifacts and increase the sensitivity for abnormalities in previously challenging areas; however, these sequences may be associated with a significant increase in scanning time.

FIGURE 19-4 Axial trace diffusion image showing susceptibility around an arteriovenous malformation. This lesion had been embolized previously. There is significant signal loss within the embolized bed (asterisks) and a halo of high signal distortion around the margin of the embolized nidus (arrows). Note that posteromedially, where a small amount of patent nidus remains, one sees low signal flow voids but considerably less susceptibility and distortion.

Physics

DTI, a refinement of DWI, characterizes the dominant vector of water motion within voxels. In contrast to DWI, where typically only 3 directions (motion-probing gradients) are obtained, DTI requires at least 6 directions to resolve the mathematical ambiguities related to oblique fiber trajectories. Some centers increase signal either by acquiring multiple acquisitions in these 6 directions or by acquiring 25 or more directions; however, with both techniques, the additional scan time increases the risk for patient motion during the sequence and subsequent misregistration of the data.

As mentioned in the discussion on DWI, the water motion illustrated on diffusion sequences is predominantly extracellular, and the vector of motion tends to parallel the white matter tracts. In DTI, the acquisition of at least six directions allows mathematical construction of a tensor ellipsoid whose major axis points in the direction of the dominant fiber tract within a voxel.¹⁷ In voxels in which white matter tracts are homogeneous and nearly colinear, this ellipsoid is thin and elongated in a so-called prolate or cigar-shaped configuration, with the dominant axis (major eigenvector) being significantly greater than the two perpendicular axes (minor eigenvectors). Such ellipsoids are anisotropic; that is, they represent voxels whose fibers have a strong directional bias. When white matter tracts are not as colinear (i.e., when there are many crossing or divergent fibers) or in areas of gray matter or CSF, the tensor ellipsoid is nearly spherical. These ellipsoids are further characterized as isotropic; that is, they lack significant directional bias.

Once tensors have been calculated, they may be presented for interpretation in numerous ways. A mean diffusivity map portrays average molecular motion that is independent of tissue directionality, so voxels with high motion such as water will be bright and voxels with low motion such as gray matter will be dark. A fractional anisotropy (FA) map represents the colinearity and integrity of fibers, and therefore large tracts with parallel white matter bundles, as seen in the callosum fibers and descending corticospinal tract, appear white (FA approaches 1). CSF, gray matter, and crossing white matter tracts all lack parallel axonal configuration and thus appear dark or black (FA approaches 0). Color-encoded directional maps overlay fiber directional data on an FA image, with typical color assignments of red representing transverse fibers, green representing anteroposterior fibers, and blue representing superoinferior fibers; brighter or more saturated fibers indicate greater colinearity and hence greater anisotropy (Fig. 19-5).

FIGURE 19-5 Normal diffusion tensor maps. A, Color-encoded diffusion tensor map showing primary tract orientations. By convention, red indicates right-left fibers, green indicates anterior-posterior fibers, and blue indicates superior-inferior fibers. Oblique directions are represented by intermediate colors, so for example, as splenial fibers leave the midline and angle toward the occipital lobe, they change from red to orange to yellow to lime-green to green. B, Fractional anisotropy map indicating the degree of anisotropy within given voxels. The voxels containing nearly colinear fibers (e.g., genu, splenium, posterior limb of the internal capsule) appear bright, with fractional anisotropy approaching 1. The gray matter has proportionately fewer fibers that are more arbitrarily oriented; they appear dark and have fractional anisotropy approaching 0. Note that when fiber tracts cross, for example, lateral to the posterior margins of the ventricles, signal drops completely because conventional acquisition and processing techniques are unable to resolve discordant orientation data.

Tractography is a mathematical method of tracking theoretical fiber trajectories by using vector data to incrementally advance to adjacent voxels. Tracking is typically initiated by placing a source “seed” over an area of interest. A second “target” seed can be placed to determine a destination focus to which the trajectory must run, or the second seed may be omitted to allow tracking software to course spontaneously to determine all potential tracts leading from the source seed.

Clinical Uses and Applications

DTI can be a very useful tool for demonstrating the integrity or compromise of tracts running between different areas of the brain. Four classic patterns of abnormal fiber orientation have been described: (1) deviated but otherwise preserved fiber tracts, (2) edematous tracts with diminished FA but preserved fiber tract orientation, (3) infiltrated tracts with diminished FA and abnormal fiber tract orientation, and (4) frank destruction of fiber tracts with negligible anisotropy.¹⁸ Figures 19-6 and 19-7 illustrate varieties of tract disruption.

FIGURE 19-6 Infiltrating parietal glioma. A, Axial T1-weighted image showing a slightly heterogeneous low-signal mass within the left parietal white matter (arrows), with the mass causing effacement of the overlying sulci. The intraparietal sulcus is compressed and displaced anteriorly (black arrowhead). The normal right intraparietal sulcus is shown for reference (white arrowhead). B, Fluid-attenuated inversion recovery T2-weignted image showing high signal corresponding with low intensity on T1 (arrows), and there is little enhancement after gadolinium (C) (arrows) . D, Fractional anisotropy map showing loss of anisotropy corresponding to tumor (arrows). E, Tensor color map demonstrating anterior displacement and more strongly inferior (more blue) orientation of fibers within postcentral gyral white matter (black arrowhead) than in the normal position on the contralateral side (white arrowhead). Note that the normal white matter anatomy seen posteriorly on the right (hatched arrows) is obliterated on the left. F, Diffusion tractography showing displacement of fibers around the tumor (arrows).

FIGURE 19-7 Diffusion tensor characterization of white matter integrity around a glioma centered near the sylvian point. A, The lesion itself has low anisotropy (asterisk) because of its high cellular nature and lack of organized white matter tracts. The surrounding white matter is splayed and compressed (arrows). As the compressed arcuate fasciculus extends posteriorly around the mass, the color orientation changes from green (anteroposterior) to yellow (anteroposterior oblique) to orange (transverse oblique) to red (transverse); as these fibers project anteriorly again, the colors repeat in reverse. B, Tractography demonstrates that the corticospinal tracts are deviated medially (arrows) but appear grossly intact.

Such assessment of tracts can aid in planning treatment and determining prognosis, thereby allowing surgeons and radiation oncologists to protect tracts that are preserved while more aggressively addressing tracts that have already been destroyed. This has tremendous surgical importance because with these data a surgeon can maximize resection while minimizing risk to adjacent structures. A study by Kikuta and colleagues involving patients who had undergone resection for arteriovenous malformations (AVMs) found that incomplete tractography of the optic radiation was associated with visual field loss postoperatively.¹⁹ Such information not only assists in surgical planning but also helps prepare patients for potential outcomes of surgery. In addition, for patients undergoing surgery to treat brain tumors near the motor pathways, tractography may be used to identify initial sites for electrocortical stimulation, thus facilitating faster localization of eloquent cortex during surgery.

Applications of DTI extend beyond tumor evaluation. Preliminary studies suggest that DTI may be sufficiently sensitive to detect early changes in vulnerable regions in individuals with certain cognitive disorders, even at the presymptomatic or preclinical stages.²⁰ Additional uses for tractography include assessment of patients with multiple sclerosis, and early studies in stroke patients suggest that DTI can aid in prognosis and may detect corticocortical rewiring.

DTI and tractography have recently been applied to the spinal cord, an area previously too degraded by artifact to image successfully. Some authors now suggest that FA data from DTI may be useful in distinguishing surrounding cord edema from tumor.²¹ In spinal cord AVMs, FA values in surrounding cord may improve after embolization and correlate with enhanced patient outcome.

Pitfalls and Limitations

A major limitation of conventional DTI is its poor distinction of crossing fiber tracts. When a voxel contains a homogeneous population of similarly oriented fibers, calculation of the dominant vector or water motion is straightforward. However, when a voxel contains crossing fibers, a simple tensor calculation is unable to reflect the more complex fiber configuration. In this instance, the major eigenvector may be calculated as the average of all voxel fibers, and as a result the eigenvector may not represent any of the dominant tracts.²² Newer techniques such as high–angular resolution diffusion imaging (HARDI) and Q-ball imaging may use 100 or more directions along with complex mathematical techniques to better characterize intravoxel tract ambiguity.²³

The DTI sequence is also prone to artifacts that result in local field distortion and sometimes signal dropout. Thus, the diffusion signal may be compromised in regions adjacent to postoperative air, blood products, and surgical clips or embolic material. As in DWI, eddy currents can also distort images in DTI and result in image misregistration and computational errors; in these instances, correction software can compensate for at least some of the distortion to yield more accurate maps.

Tractography depends on user input to define tracts of interest. Incorrect or suboptimal region-of-interest placement will result in misrepresented tracts. Mathematical assumptions will also influence tractography generation, with tract validity being dependent on appropriate definition of the maximal angulation within a voxel, the minimum FA value to tolerate while generating the tract, and the total path length. Because signal in the cord is more variable than signal in the brain and because the cord moves with CSF pulsation, spinal tractography proves considerably more challenging. In general, tractography in the cervical cord is easier to generate than in the thoracic cord, at least in part because of the greater pulsation effects in the mid and lower cord.²¹

Physics

TOF MRA is performed by applying repetitive pulses to stationary tissues in a discrete volume of the brain, which results in saturation or suppression of signal in these tissues. Blood flowing into the volume has not been saturated and is therefore fully magnetized; this in-flowing blood therefore provides the only significant signal within the imaging slab. The intrinsic contrast that is derived from flowing blood eliminates the need for injection of contrast material to visualize the intracranial vessels. The pulse sequence used typically consists of a gradient recalled echo (GRE) sequence, which is acquired with either a two-dimensional (2D) or 3D technique. 2D TOF images are obtained as contiguous or slightly overlapped sections, whereas 3D TOF images are derived from one or more overlapping 3D volumes. Of these two methods, 3D TOF is more commonly used than 2D TOF because of its superior spatial resolution.²⁴

In PC MRA, the vascular contrast is obtained by applying a bipolar phase-encoding gradient and a velocity-encoding (VEnc) factor.^25,26 Phase shifts in moving spins or flowing blood are obtained when gradients with opposing polarities are applied twice during a single RF excitation. This method results in nulling of signal from stationary tissue while exploiting signal from flowing blood. Visualization of blood flow depends on the velocity of the flow. In specific encoding orientations, middle-gray usually represents no flow, whereas progressively increasing shades of white or black indicate directionally encoded higher velocities up to the predefined VEnc factor. Blood that flows at velocities higher than the VEnc factor will demonstrate aliasing, a situation in which signal intensity “wraps around” and will therefore be opposite that expected. In other words, if flow in a certain direction matches the VEnc factor, it will be encoded as white, but any greater velocity in that direction will actually appear as black. Because arterial flow is more rapid than venous flow, the typical arterial VEnc factor intracranially measures greater than 60 to 80 cm/sec, whereas slow flow within veins and venous sinuses is best imaged with a VEnc factor of 20 cm/sec or less.²⁶ Images generated from this technique provide not only direction of flow but also magnitude of flow.

CE MRA uses a high-resolution T1-weighted technique in which the vessels are accentuated by a bolus of gadolinium chelate. The 2% to 3% concentration of gadolinium contrast agent in blood (by volume) causes a marked T1-shortening effect in vessels with a resultant increase in vessel signal. CE MRA provides anatomic information related to lumen diameter and the concentration of contrast material in vessels rather than physiologic information reflecting the flow rate, as calculated with TOF and PC techniques. Advantages of CE MRA over non-CE techniques include higher signal-to-noise ratios, decreased susceptibility to artifacts caused by pulsatility and flow, shorter image acquisition times, and temporal resolution that illustrates patterns of flow over time. Older CE MRA paradigms (e.g., elliptic-centric techniques) had collected full RF spectral data during the bolus, so only about three full-volume MRA repetitions could be acquired per minute. However, with that technique, the intended arterial-phase imaging was frequently obscured by venous enhancement. The need for improved temporal resolution led to the development of time-resolved CE MRA, such as TRICKS (time-resolved imaging of contrast kinetics) and TREAT (time-resolved echo-shared angiography technique).²⁴ These sequences improve temporal resolution by repetitively acquiring the center of the RF spectrum along with varying peripheral portions during each acquisition; data missing from unsampled regions are interpolated by using data obtained at other points in time. This development enables more rapid 3D acquisitions before, during, and after transit of a contrast bolus, thereby allowing identification of the arterial, capillary, and venous phases.^27,28

Clinical Uses and Applications

The angiographic modality with unparalleled spatial and temporal resolution is DSA. However, manipulation of catheters in the aorta or in the carotid or vertebral arteries, even by an experienced operator, is associated with a 1.3% risk for a cerebral event.²⁹ Therefore, in clinical practice most institutions still use noninvasive techniques such as computed tomographic angiography (CTA) or MRA for preliminary evaluation of the vascular tree in the clinical setting of infarction, aneurysm, AVM, dural arteriovenous fistula (AVF), and veno-occlusive disease.

Intracranial Aneurysms

Detection of intracranial aneurysm can be done with noninvasive techniques such as MRA and CTA. 3D TOF MRA has been shown to have an overall sensitivity of 87% and specificity of 95% for the detection of intracranial aneurysms (Fig. 19-9). The sensitivity for detection of aneurysms larger than 3 mm is greater than that for aneurysms 3 mm or smaller (94% versus 38%) on older 1.5T scanners.³² At a magnetic field strength of 3T, aneurysms as small as 1 mm can be detected.³³ Both TOF MRA and CE MRA are optimized to show vessel lumens; apical thrombus may remain undetected unless MRA is correlated with conventional sequences that are more sensitive for nonflowing blood.

FIGURE 19-9 Surface rendering of a three-dimensional time-of-flight magnetic resonance angiogram reveals an aneurysm at the left internal carotid artery bifurcation (arrow).

Studies have shown similarities and differences between 3D TOF MRA and CE MRA in the detection of aneurysms. One study demonstrated no significant difference in the quality of the images and equal detection rates for both techniques.³⁴ Another study showed that 3D TOF MRA detected more aneurysms than CE MRA did,³⁵ whereas a third study indicated a sensitivity of 100% and a specificity of 94% for CE MRA, hence suggesting that CE MRA may be superior to the 3D TOF technique.³⁶ Thus, there remains no clear consensus regarding which modality is superior, but both provide relatively accurate noninvasive methods for the detection of aneurysms.

MRA has also been studied for the follow-up of intracranial aneurysms treated by coil embolization. Serial studies have been performed with the use of both 3D TOF and CE MRA after treatment to evaluate for delayed aneurysm configuration because recanalization is estimated to occur in 10% to 40% of patients.³⁷ Studies evaluating residual flow in coil-treated aneurysms have shown a sensitivity and specificity of 81% and 90.6%, respectively, for 3D TOF MRA and 86.8% and 91.9%, respectively, for CE MRA.³⁸ It has been suggested that CE MRA may be slightly more sensitive in evaluating residual slow flow.²⁴ DSA does remain the gold standard for evaluation of residual filling in treated aneurysm, but studies have shown 86% to 94% agreement between 3D TOF MRA evaluation and DSA.^39–41

Vascular Malformations

AVMs and AVFs are typically evaluated with DSA because of its excellent spatial and temporal resolution. However, MRA techniques may also provide an accurate noninvasive method for preoperative planning and evaluation of AVMs.^42,43 3D TOF MRA and PC MRA allow characterization of flow in abnormal vessels, but they do not provide information on the hemodynamics of vascular malformations.⁴⁴ Because definition of the architecture in and around malformations (e.g., arterial feeders, nidus size, and draining veins) is crucial in the evaluation of AVMs, advances in temporal resolution have had a large impact on noninvasive evaluation of these abnormalities. Indeed, ultrafast CE MRA and time-resolved CE MRA techniques have been shown to define relationships between lesions and the vasculature better than TOF and routine CE MRA techniques.^45,46 Specific types of time-resolved CE MRA such as TRICKS MRA have been developed to provide hemodynamic information that gives one the ability to detect processes previously undetectable on noninvasive imaging, such as early venous drainage (Fig. 19-10).²⁴ For example, in a small series of 40 patients, time-resolved MRA at 3 T was shown to be a reliable technique for the evaluation and surveillance of dural AVFs.⁴⁷

FIGURE 19-10 Time-resolved contrast-enhanced magnetic resonance angiography showing embolization of an arteriovenous malformation. All images are thick-section maximum intensity projections at progressively later points in time during the bolus transit. A-C, Pre-embolization scans at early arterial (A), slightly later (B), and discrete venous (C) phases. The corresponding postembolization images D to F parallel A to C. In the pre-embolization arterial scan (A), a large posterior cerebral artery feeder supplies an abnormal large wedge-shaped parietal-occipital area (arrows). The presence of venous enhancement in this early arterial phase (arrowheads) confirms abnormal shunting. The arterial phase in the postembolization image (D) shows a considerable reduction in the volume of the perfusing nidus, with a small focus remaining posteriorly (arrows). Note that even with the small residual nidus there is still early venous enhancement (arrowheads).

Intracranial Venous System and Sinus Venous Thrombosis

MRA techniques can be applied to study the intracranial venous system. Although arterial flow provides robust signal on 3D TOF sequences, the slower flow in veins provides more limited signal with 3D techniques. 2D TOF techniques are used instead because they offer excellent sensitivity for slow flow and less saturation effects than with 3D TOF techniques (Fig. 19-11).⁴⁸ 2D TOF best depicts flow perpendicular to the scanning plane, so areas of turbulent or tortuous flow (e.g., from the straight sinus through the transverse sinus through the sigmoid sinus) may require imaging in more than one acquisition plane to adequately represent flow patterns. In addition to TOF magnetic resonance venography (MRV), 2D or 3D PC MRV may be used because these techniques can indicate not only the rate of flow but also the directionality of flow. However, the dependence of PC MRV on operator-defined VEnc parameters may complicate its use,⁴⁹ with incorrect assumptions at the time of scanning potentially leading to artifacts on the final images.

FIGURE 19-11 Dural venous thrombosis. A, Sagittal paramidline non–contrast-enhanced T1-weighted image demonstrating increased signal in the transverse sinus (arrow), suggestive of thrombosis rather than flowing blood. B, Coronal source data from two-dimensional time-of-flight magnetic resonance venography posteriorly shows low signal where the superior sagittal sinus is expected (arrow), but flow within adjacent cortical veins is well seen (arrowheads). C, More anteriorly, flow is confirmed in the superior sagittal sinus (arrow), which drains anteriorly. D, A sagittal maximum intensity projection of time-of-flight venography data better defines the flowing superior sagittal sinus anteriorly (hatched arrows) but not posteriorly (arrows). Flow within cortical veins (arrowheads) should not be mistaken for patency of the superior sagittal sinus.

MRV can be enhanced by exploiting the paramagnetic effects of intravenous gadolinium. CE MRV has been shown to provide significantly better visibility of venous structures, including deep cerebral veins, than possible with the TOF technique. Selected venous structures studied in a small group of patients demonstrated visibility of 99% with CE MRV and 72% with TOF MRV. CE MRV can also be helpful in reducing the effects of turbulent flow on visualization of the venous sinuses.⁵⁰

Pitfalls and Limitations

Various MRA techniques can serve as an alternative to DSA imaging, although limitations in both spatial and temporal resolution still render DSA the gold standard in evaluating most neurovascular pathologies. However, MRA is used widely because it is noninvasive (or minimally invasive if gadolinium is injected). Therefore, it is important to be familiar with limitations of the various vascular MR techniques.

Spin dephasing can occur when blood flow is complex or turbulent and when vessels are in close proximity to tissues with short T1 properties, such as fat and subacute hemorrhage (methemoglobin). This configuration may result in loss of signal on TOF and PC MRA. Slow flow can also result in similar loss of signal on these imaging sequences because of spin saturation.^51,52 Flow saturation effects can be diminished by using the “multiple overlapping thin slab acquisition” (MOTSA) technique for 3D TOF MRA (Fig. 19-12). Most institutions use three or four such slabs in an MRA acquisition through the head; this technique improves flow signal in vessels but lengthens imaging times.

FIGURE 19-12 Coronal maximum intensity projections of a three-dimensional time-of-flight magnetic resonance angiogram (A) and segmented projection of the vertebrobasilar system only (B). The sequences are ordinarily acquired by using multiple overlapping slabs to increase the conspicuity of vessels. Where slabs overlap (indicated schematically by bright green reflecting slabs 1 plus 2, or 2 plus 3, or 3 plus 4), one sees better signal but also abrupt increase in venous signal (arrows in B).

Velocity aliasing can be seen with PC MRA when true velocities exceed the VEnc parameter of the imaging sequence.⁵³ This may result in the perception of motion opposite the actual direction of blood flow; however, experienced radiologists will recognize the abrupt transition from black to white (or vice versa) as artifactual. In addition, factors such as lengthier acquisition times, lower spatial resolution, and artifacts related to pulsatile flow have limited the use of PC MRA in comparison to TOF and CE MRA.⁵⁴ Thus, many sites reserve PC MRA to answer specific questions about flow directionality rather than including it as a standard imaging sequence.

Studies have reported high sensitivity and specificity ranging from 94% to 96% for 3D TOF MRA in the evaluation of intracranial stenosis. However, the MRA technique has a tendency to overestimate the degree of stenosis. A small study demonstrated overestimation of stenosis in 37% of diseased intracranial arterial segments with 3D TOF MRA when compared with DSA.⁵⁵ Even though such measurement errors do limit the positive predictive value of TOF MRA in evaluating intracranial stenosis, this technique is still frequently used because of its high negative predictive value and its noninvasive technique.⁴⁴

Failure to visualize appropriate signal in a normally flowing sinus may relate to MRV acquisition in a plane parallel to that of venous flow. This is a common limitation in the interpretation of these images and often mandates closer assessment of the source images. In instances in which true thrombosis is suspected, the acquisition plane may be modified, or CE MRV may be performed to improve sinus visualization.⁴⁹

Physics

As in PC MRA, a PC acquisition can be used to quantify the rate and directionality of CSF flow. A similar bipolar gradient is applied to stationary tissue to cancel the magnetization from stationary spins and null the signal from these tissues. CSF that moves through the plane of acquisition between two separate gradient pulses will experience a net phase shift that is proportional to its velocity along a selected axis.^56,57 The VEnc factors used in PC CSF studies are typically lower than those used for PC MRA, averaging 10 to 15 cm/sec. Imaging is usually acquired in the midline sagittal plane, although other orthogonal or oblique planes can be prescribed as well if needed.

PC CSF study differs from PC MRA in its use of cardiac gating. Hence, with this technique, magnitude and direction of flow can be segregated by phase in the cardiac cycle. Sixteen phases are typically acquired through the cardiac cycle, with triggering generally based on detection of each electrocardiographic R wave.⁵⁸ In patients with normal sinus rhythm, adequate data are acquired through the cardiac cycle. Certain arrhythmias may compromise gating, however, and limit signal, particularly in the later phases.

Clinical Uses and Applications

PC MRI has been used to evaluate CSF flow in a wide range of cerebral disorders, including communicating hydrocephalus, normal-pressure hydrocephalus, and aqueductal stenosis. Recently, analysis of CSF pulsation in the axial plane at the foramen magnum has improved our understanding of the role of CSF flow disturbances in patients with symptomatic Chiari I malformations. PC MRI has been used to demonstrate elevated peak systolic velocities through the subarachnoid space and foramen magnum in these patients in comparison to normal subjects.⁵⁹

Normal Cerebrospinal Fluid Flow

PC MRI sequences are usually analyzed in both static and cine formats. Typically, CSF flow sequences will generate magnitude, phase, and three orthogonal PC flow-encoded sets. Of greatest utility are the directional series that characterize flow within a structure. In the case of midline CSF flow from the third ventricle through the fourth ventricle outlet foramina, the most useful direction would conform most closely to the z-axis (Fig. 19-13). Shades representing one direction versus another may vary by scanner vendor; in practice, radiologists typically refer to the structures with known flow patterns (e.g., the deep venous system) to determine which shades represent which directions of flow.

FIGURE 19-13 Normal cerebrospinal fluid (CSF) flow pattern. A superior-inferior sensitized image from a phase-contrast CSF flow study shows flow through the aqueduct (arrow), through the fourth ventricle, and past the outlet foramina (hatched arrow).

During a normal cardiac cycle, there is rhythmic movement of not only CSF but also brain parenchyma. Cardiac pulsations are transmitted through the intracranial arteries and capillaries and result in motion within the CSF. The relationship between the cardiac cycle and CSF is perceived as normal oscillatory CSF flow. During systole, expansion of the cerebral arterial vessels creates a pressure wave within the subarachnoid CSF. This pressure wave is transmitted through the intracranial subarachnoid space and results in outflow of CSF through the foramen magnum into the compliant spinal CSF space. In diastole, the arterial relaxation and resulting change in intracranial pressure cause CSF to flow back into the cranium through the foramen magnum.^57,60 PC MRI techniques can be used to demonstrate the pulsatile flow within the CSF compartments and have been used to demonstrate the mechanical coupling between blood flow and CSF flow.^61–64 Cine CSF flow studies have shown that normal CSF pulsations exhibit a timed sequence of oscillatory flow at specific sites in the ventricular system and subarachnoid spaces.⁵⁸ The full length of the aqueduct can often be outlined by upward and downward waves of CSF pulsation (Fig. 19-14). CSF flow within the aqueduct is directed caudally during systole and cranially during diastole with an approximate flow velocity of 3 to 5 cm/sec.⁶⁵

FIGURE 19-14 Color-encoded superior-inferior sensitized images from a phase-contrast cerebrospinal fluid (CSF) flow study. A, Earlier in the cardiac cycle there is predominantly inferior flow (blue) through the aqueduct and beyond the outlet foramina. Some inferior flow is also seen in the prepontine cistern as well, with portions of the superiorly flowing basilar artery represented in red. B, Later in the cycle there is predominantly superior CSF flow anterior to the brainstem (in red). Some turbulent flow is seen in the posterior third ventricle (mixed red and blue) from retrograde pulsation through the aqueduct.

Abnormal Flow of Cerebrospinal Fluid

Alterations in normal CSF flow dynamics can be seen in a variety of neurological conditions. Normal oscillatory CSF can be affected when soft tissue impinges on or impacts the foramen magnum, as seen in symptomatic Chiari I malformations. Abnormal flow can also occur in patients with Dandy-Walker malformations, arachnoid cysts, or fourth ventricular outlet obstruction. CSF pulsation waves in these cases appear confined to the isolated space without detection of any significant CSF flow across the point of obstruction (Fig. 19-15).⁵⁸ A mass that partially obstructs CSF outflow may also alter the time course of CSF pulsations.

FIGURE 19-15 Four-year-old girl with septo-optic dysplasia and colpocephaly who had previously been shunted. A, Anatomic sagittal T1-weighted image showing the markedly enlarged ventricular system with flattening of the third ventricle, and relative sparing of the aqueduct of Sylvius, and fourth ventricle. B, Superior-inferior sensitized image from a phase-contrast cerebrospinal fluid flow study showing some pulsation through the posterior third ventricle (faint dark triangle projecting anterior and superior to the arrowhead) but no flow through the aqueduct itself (arrowhead and below). Flow may not be seen through even patent aqueducts in shunted patients.

An imbalance between intracranial blood and CSF flow has been suggested to be responsible for certain neurological disorders such as normal-pressure hydrocephalus.⁶⁶ PC MRI has provided a means for the measurement of CSF flow at the aqueductal level. This has allowed improvement in diagnosis and management of patients with communicating hydrocephalus.^67–69 It has been shown that a small group of symptomatic patients with CSF stroke volume greater than 42 µL per cycle may benefit from CSF shunting⁶⁹; in normal volunteers the aqueductal stroke volume averages 39.51 ± 18.21 µL per cycle (range, 8.68 to 71.7).⁷⁰ Quantification of the peak flow rate has likewise been suggested as an appropriate means for detecting abnormal aqueductal flow.⁷¹

The use of PC MRI has also had important implications in the management of aqueductal stenosis or compression and the management of posterior third ventricle masses. The diagnosis of impaired flow through the aqueduct as the cause of hydrocephalus can direct management to endoscopic third ventriculostomy, which has shown success in the treatment of obstructive hydrocephalus.^72,73 Although conventional MRI may demonstrate signs of aqueductal stenosis such as triventricular dilation or downward bulging of the third ventricle, evaluation with these signs alone is subjective and highly dependent on observer experience.⁷⁴ PC MRI offers a reliable, reproducible, and accurate method for the diagnosis of aqueductal compromise in patients with suspected obstructive hydrocephalus by demonstrating absence of CSF flow (Fig. 19-16).⁷⁵

FIGURE 19-16 Representation of successful third ventriculostomy. A sagittal superior-inferior sensitized image from a phase-contrast cerebrospinal fluid (CSF) flow study (A) and a sagittal T1-weighted fluid-attenuated inversion recovery image (B) confirm to-and-from motion of CSF through the ventriculostomy in the floor of the third ventricle (arrows), with a resulting jet into the third ventricle itself (arrowheads).

Pitfalls and Limitations

The accuracy of CSF flow evaluation can be limited if the proper VEnc factor is not correctly selected by the operator. If CSF velocity exceeds the preselected VEnc parameter, aliasing or phase wraparound may occur. It is important to understand that the information acquired with PC MRI techniques without the use of true cardiac gating is an approximation of true CSF flow. The images obtained do not represent specific data points within the cardiac cycle but rather are approximations of preselected intervals with respect to the heart rate.⁵⁸

Knowledge of certain pitfalls in PC MRI can greatly enhance diagnostic accuracy. For example, absent CSF pulsation may not indicate aqueductal obstruction; in some situations slow flow of CSF in general (as with obstruction at the level of the foramen of Monro by a colloid cyst) may result in reduced CSF flow downstream that could be interpreted as impaired flow through the aqueduct.⁶³ In addition, CSF pulsations can appear to be discontinuous in a patent CSF space when there is converging flow, turbulence, or aliasing of pulsations.⁵⁸

Physics

Perfusion examinations are used to make relative and quantitative estimations of the blood flow that supplies different elements or regions of the brain. As in computed tomographic (CT) perfusion imaging, the main hemodynamic values generated by MR perfusion studies are cerebral blood flow (CBF), cerebral blood volume (CBV), and mean transit time (MTT). Hemodynamic perfusion MRI studies can be performed either with or without intravenous contrast enhancement, but most clinical perfusion performed today uses a concentrated bolus of gadolinium to modify signal within the tissue as the bolus perfuses.

Dynamic susceptibility contrast perfusion-weighted imaging (PWI) exploits changes in T1 and T2 tissue contrast during a gadolinium bolus. For brain tumor imaging, a standard gadolinium dose (0.1 mmol/kg of body weight, typically corresponding to 15 to 20 mL of contrast agent) will provide an excellent signal-to-noise ratio with GRE EPI.⁷⁶ Typical injection rates typically range from 2.5 to 5.0 mL/sec, and many sites use a double-barreled injector to flush saline after gadolinium to maintain a tight contrast bolus. The brain volume is scanned repeatedly before, during, and after the bolus, perfusion, and a time-intensity curve is generated for each voxel (Fig. 19-17). In general, the area under the curve correlates with CBV, the maximum slope of the curve roughly approximates CBF, and MTT refers to the width of the susceptibility curve. These three parameters are related by the equation CBF = CBV/MTT.⁷⁷ It is worth noting that unlike CT perfusion, in which image density is linearly dependent on contrast concentration, in MR perfusion the relationship of signal to contrast concentration is curvilinear. As a result, most contrast MR perfusion studies in the brain will refer to relative CBV and relative CBF, and therefore comparison of suspected pathologic perfusion with normal contralateral perfusion is necessary.

FIGURE 19-17 Relative cerebral blood volume map of a patient with a large necrotic glioma. The area of suspected viable tumor has variable blood patterns. Centrally, where necrosis was identified on anatomic imaging, there is little or no blood volume (asterisk). A thin band of increased blood volume is seen along the medial aspect of the cavity (arrowheads), but it is the lateral aspect of the cavity (arrows) that shows the most prominent perfusion pattern. Characterization of such blood flow patterns may aid surgeons by identifying the most metabolically active areas for biopsy planning so that the pathologist can grade the most aggressive portion of the tumor.

A less prevalent variety of dynamic scanning is permeability imaging (Fig. 19-18). Although gadolinium does not normally pass through the blood-brain barrier (BBB), when the BBB is disrupted there is slow extravasation of contrast material into surrounding tissue. The pattern and rate of extravasation can be documented by dynamic repetitive acquisition of rapid T1-weighted MR images. This technique enables calculation of the microvascular permeability of the BBB. Permeability may be followed by perfusion imaging in one study session, with the added advantage that the permeability scan pre-enhances pathologic tissue and thus results in more normal return of the perfusion curve to baseline during PWI.

FIGURE 19-18 Permeability map in the same patient as in Figure 19-17. A, Axial T1-weighted gadolinium-enhanced image showing region-of-interest placement and (B) corresponding time-dependent enhancement curves. Curve 7 from the more anterior cavity shows a rapid first-phase rise and continued rise in the second phase at a lower rate. This is suggestive of radiation necrosis. Curve 8 from the necrotic cavity more posteriorly shows a rapid rise in the first phase but leveling out in the second phase. This is suggestive of viable tumor.

Arterial spin labeling (ASL) does not use intravenous gadolinium but instead assesses perfusion by labeling protons in the large arteries of the neck.^78,79 These labeled protons then course intracranially within the blood pool and result in diminished signal. A second acquisition is generated without water labeling to serve as a baseline volume. Hemodynamic parameters and maps are generated on the basis of differences between the two acquisitions (Fig. 19-19).⁸⁰

FIGURE 19-19 Tumor recurrence. A, Axial T1-weighted contrast-enhanced image revealing a nodular area of enhancement in the right superior temporal lobe (arrow). B, Relative cerebral blood volume map showing a very subtle nodule of increased perfusion (arrow) that could escape detection among the many sylvian vessels. C, Continuous arterial spin labeling map showing the focus of increased perfusion (arrow) more clearly than possible with the gadolinium bolus technique. In both B and C there is generalized diminished perfusion to the right hemisphere, particularly posteriorly, from previous radiation therapy. D, A colorized spin labeling image from C superimposed on the anatomic image from A confirms that both abnormalities (arrow) represent the same process.

Another technique used for obtaining perfusion images without gadolinium enhancement depends on the BOLD technique, which is discussed further in the section “Functional Magnetic Resonance Imaging.”

Clinical Uses and Applications

The combination of PWI and DWI in patients with brain ischemia can help distinguish viable tissue that is at risk for infarction (i.e., the penumbra) from the nonviable infarcted tissue core, which would have DWI restriction and markedly reduced CBV (Fig. 19-20).⁷⁷ In addition to its application in cerebral ischemia from large-artery stenosis, as well as small-vessel disease, PWI can be used to characterize changes in perfusion in such entities as vasculitis, moyamoya, and vasospasm (Fig. 19-21). This holds particular promise in the assessment of peripheral ischemia in the setting of vasospasm (as from subarachnoid hemorrhage), which is difficult to quantify with catheter angiography.

FIGURE 19-20 Diffusion-perfusion mismatch. A, An axial diffusion-weighted image obtained within 8 hours of the onset of symptoms reveals a large focus of restricted diffusion primarily conforming to the left lenticular nucleus, with additional involvement of the caudate head anteromedially and the anterior and posterior limbs of the internal capsule (arrowheads). This was dark on the apparent diffusion coefficient map and corresponded with acute infarction. B, A dynamic susceptibility perfusion imaging cerebral blood volume map shows variable patterns within the infarct bed; the spectrum ranges from very low blood volume in the posterior lenticular nucleus (black asterisk), to slightly higher volume in the mid and anterior lenticular nucleus (white asterisk), to an increased vascular pool (greater than the contralateral side) in the left caudate head (hatched arrow). There is a generalized increased volume in the more superficial left middle cerebral artery territory (arrows). C, The mean transit time map parallels the blood volume map, with prolonged transit time through the penumbra, but the posterior lenticular infarct (black asterisk) and internal capsule have slightly more rapid transit than does much of the remaining infarct bed (white asterisk). D, Delayed axial nonenhanced computed tomography shows that the final infarct territory (arrowheads) correlates well with the diffusion abnormality; the more superficial penumbra did not ultimately infarct.

FIGURE 19-21 Sixty-five-year-old patient with a history of coronary heart disease evaluated for acute left upper extremity weakness. A, Focus of restricted diffusion in the right precentral gyrus (arrow) was confirmed on an apparent diffusion coefficient map. B, A frontal maximum intensity projection of a 3D TOF MRA through the right internal carotid artery and tributaries confirms the stenosis (arrow), with irregular calibers of the more distal branches. C, Phase-contrast magnetic resonance angiography, with left flow indicated in white and right flow indicated in black, demonstrates a focal narrowing in the luminal profile (arrow) and poor flow more distally. D, Region-of-interest intensity curves over the central white matter in each hemisphere (right, green; left, purple). The more gadolinium that perfuses a vessel, the greater the drop in signal. E, Smoothed curves showing the slight delay in perfusion to the right (the green arrowhead indicates the later peak). The area under the curve, bounded by the curve itself and a line drawn between the shoulders of the curve, is proportional to blood volume. F, The resulting blood volume map shows increased volume to essentially the entire right middle cerebral artery territory. Vasodilation is an early autoregulatory responses to diminished pressure.

Perfusion examinations are also frequently performed to estimate the hemodynamics of brain tumors. Perfusion data can aid in the differential diagnosis, with increased perfusion seen in gliomas as a result of angiogenesis related to aggressiveness and tumor viability.⁸¹ PWI has also become useful after treatment in distinguishing radiation necrosis from tumor recurrence.⁸² Both tumor growth and radiation necrosis will exhibit contrast enhancement related to disruption of the BBB, but the enhanced perfusion of neoplastic tissue (similar to or greater than adjacent “normal” tissue) contrasts with the diminished perfusion in radiation necrosis. Identification of zones of hyperperfusion may also help the surgeon target the most metabolically active area if the patient requires subsequent biopsy.⁸³

Pitfalls and Limitations

As a technique optimized to image susceptibility, dynamic susceptibility contrast imaging is also exquisitely sensitive to static structures in and around the brain that induce a strong magnetic field inhomogeneity. Thus, chronic blood products (hemosiderin), calcification, metal, or air (postoperative or in a paranasal sinus/mastoid cell) can cause artifacts and distortion of the perfusion data set. At standard slice thicknesses and gaps, small regions of hypoperfusion may escape detection, and attempts to increase resolution will reduce signal and prolong the scan time with a resulting reduction in temporal resolution.

With regard to tumors, distinction between tumor growth/recurrence and radiation necrosis can still be confusing and problematic, especially since both processes may coexist in similar areas.⁸⁴ In addition, calculation of relative CBV can be grossly inaccurate in lesions such as glioblastoma multiforme, in which there is severe breakdown of the BBB.⁷⁸ Furthermore, the extent of tumor infiltration into adjacent tissue is suboptimally characterized, particularly in areas where there is early infiltration but not yet neovascularity. Finally, tumor histology cannot reliably be inferred from perfusion data because some of the lower grade cerebral tumors such as oligodendrogliomas can have markedly increased perfusion that would otherwise be more suggestive of higher grade tumors.

Physics

Functional MRI (fMRI) characterizes brain hemodynamic activity by indirectly revealing areas of neuronal activation after various stimulation by motor, visual, auditory, and tactile sensory input, most commonly in the form of “block paradigms.” Shortly after presentation of stimuli (or the onset of motor activity), the increase in cortical activity leads to increase in local CBF. The oxygen delivered by this increased blood flow is greater than the demand of tissue for oxygen, and as a result there is a net increase in oxyhemoglobin in the capillary bed. This increase in oxyhemoglobin is associated with a relative decrease in deoxyhemoglobin, whose paramagnetic properties ordinarily cause a reduction in local signal. Hence, with the reduction in deoxyhemoglobin comes a transient net increase in local signal.⁸⁵ Because this technique depends on variation in oxygenation levels in blood, it is frequently referred to as BOLD imaging (Fig. 19-22).

FIGURE 19-22 Representative signal profile of an activated voxel during a functional MR task using blood oxygen level–dependent imaging. The intensity of this voxel in the hand motor cortex region is shown during a finger tapping paradigm in which the subject had five “on” periods and five “off” periods. The cyclic pattern of signal reflects changes in oxygen concentration within portions of activated brain. Almost immediately after subject begins a task, the amount of oxygen within the activated territory increases and results in a rise in signal. After the task is complete, the oxygen levels return to baseline and the voxel signal declines.

BOLD EPI therefore identifies areas of increased MR signal from activated brain cortex. In practice, typical paradigms consist of three or four sets of alternating “on” periods, where sensory stimulation, cognitive tasks, or motor activity is present, and “off” periods, which are free of such stimulation. Differences between “on” and “off” measurements help determine where blood flow is affected by the stimulus or activity. Activation maps are generated by determining the statistical threshold above which true activity is likely. This activation pattern is then typically color-coded and superimposed on a high-resolution T1-weighted MRI sequence of the same section or on a 3D model for anatomic recognition.⁸⁶

Clinical Uses and Applications

Mapping of cognitive functions has clear implications for surgical planning (Fig. 19-23). Preoperative identification may limit intraoperative awake testing if certain functions are localized away from the surgical bed. Functional data can be further exploited by processing in conjunction with DTI data; a site of activation can serve as a seed for DTI tractography to identify the trajectory of fibers (such as pyramidal tracts) that emanate from certain cortical areas so that they may be avoided during surgery.⁸⁷

FIGURE 19-23 Functional magnetic resonance imaging activation map corresponding with a right hand finger-tapping exercise, with color intensity being proportional to activation measured with a blood oxygen level–dependent technique. Activation tends to involve the sulcus adjacent to the cortex because of the prominent blood pool in the sulcus that contains draining veins.

fMRI may also be used to demonstrate the eloquent regions of the brain and to determine right/left dominance. Most studies are sufficient to demonstrate the location of sensorimotor, speech, and auditory areas of the cerebral cortex.⁸⁸

A growing research application of fMRI is for the elucidation of mechanisms of functional recovery from illness or brain injury. For example, in patients with motor deficits after stroke, fMRI may detect evidence of recruitment of brain regions or a shift in the activation center that may portend early stroke recovery.⁸⁹

Pitfalls and Limitations

One must keep in mind that the BOLD technique indirectly measures neuronal activity based on oxygenation levels. Hence, some significant sites of neuronal activity may escape detection if they do not generate an adequate hemodynamic response. Furthermore, neuronal activation is markedly faster than the resulting hemodynamic response. Therefore, if used to detect epileptogenic foci, fMRI may better demonstrate secondary or propagated foci than the true or initial focus.⁹⁰

Signal changes during BOLD imaging are dependent on magnetic field strength. It has been shown that activation of the primary motor and visual cortex at 3 T results in a 36% and 44% increase, respectively, in activation signal when compared to 1.5 T.⁹¹ This dramatic increase in signal-to-noise ratio may clarify true sites of activation at higher fields and may help in tasks that are traditionally difficult to image, such as in language paradigms and retinotopic mapping. However, with higher field strength comes greater susceptibility artifacts, so not all patients may benefit from scanning on a higher field system.

The spatial resolution of the data from fMRI is significantly less than that of a standard T1-weighted sequence, but fMRI offers high temporal resolution given the fast EPI sequences generally used. Still, uncorrected motion during a study may result in mismapping of activation on the anatomic sequence. fMRI studies do rely on patient cooperation, and some paradigms may be too difficult for impaired or sedated patients to comprehend or perform. Image quality may be further compromised as a result of miscommunication between the scanner operator and patient, which may be due to language barriers or inadequate instruction given to the patient to ensure that the paradigm is closely adhered to. Consequently, it is important to monitor patient performance during the scan to determine the applicability of the results when they are processed.

Patients with damage to certain cortical areas by tumors can produce false results. Thus for example, the regional hemodynamics about a brain tumor itself can produce areas of activation that are not truly related to the block paradigm stimuli.⁹² In addition, activation around tumors may be altered due to changes in vascular flow patterns, a phenomen known as neurovascular coupling.⁹³

Physics

Proton magnetic resonance spectroscopy (¹H-MRS, or simply MRS) is a noninvasive in vivo method of assessment of brain metabolites. MRS is based on “chemical shift,” or modification of proton resonance frequency, in certain molecular environments. Because the frequency of chemical shift varies by field strength, in clinical practice the shift is scaled to parts per million (ppm) so that peak locations are uniform across all varieties of scanners. The MRS data are ordinarily presented as a linear graph with the chemical shift (in ppm) on the x-axis and the plot of relative signal amplitude on the y-axis.⁹⁴ However, metabolite-specific color gradients may also be superimposed over anatomic data to show relative quantities of these metabolites in specific areas.

Two basic imaging methods are used, single voxel (SV) and multivoxel (MV, also called chemical shift imaging). In SV MRS, one discrete volume (voxel) of brain is chosen for the analysis, so the voxel must be placed precisely at the time of scanning for optimal results. MV MRS obtains the spectral data for a large grid of voxels simultaneously; this technique permits assessment of regional changes in metabolite concentrations, allows comparison of one voxel with another that is thought to be normal, and can help determine the most metabolically active zone to approach when a biopsy is being considered (Figs. 19-24 and 19-25).

FIGURE 19-24 Sixty-nine-year-old patient with lymphoma. A, Axial T2-weighted image with a spectroscopy grid overlying the image. The lesion is moderately hyperintense, with a more intense central component. B, Axial contrast-enhanced T1-weighted image revealing moderate slightly stippled enhancement in the posterior of the tumor (arrow) with relative sparing anteriorly. C, Multivoxel spectroscopy (TE = 144 msec) reveals prominent choline peaks (yellow arrowheads) in virtually every voxel, especially in the central part of the tumor. Prominent chlorine peaks are characteristic of lymphoma.

FIGURE 19-25 Central nervous system lymphoma. A, Sagittal T2-weighted image demonstrating increased signal within the left cerebellar corpus medullare (arrow). B and C, Multivoxel spectroscopy grid showing the location of the voxel (in green) and corresponding spectrum. The dominant peak is choline (Ch). Preservation of the N-acetylaspartate (NAA) peak suggests volume averaging of adjacent parenchyma or incomplete tumor penetration through the voxel. Cr, creatine; MI, myo-inositol; SNR, signal-to-noise ratio. D, Axial magnetic resonance permeability image confirming greatest enhancement at the voxel site (arrow).

When interpreting the MRS data it is crucial to consider what metabolites are found in normal spectra and at what concentrations. Conspicuity of metabolites is affected by the echo time (TE) used, which ranges from very short (25 msec or less) to intermediate (144 msec) to long (288 msec).

N-Acetylaspartate (NAA, 2.0 ppm) is the highest peak in normal spectra and is classically described as a marker for neuronal integrity and density. Malignant tumors and neuronal death are associated with a decline in NAA.

Choline (3.2 ppm) is a metabolic marker of membrane density and integrity. Malignant tumors show an increase in the choline peak because of increased cell membrane turnover; however, elevated choline can also be seen in inflammatory processes as well.

Creatine (3.0 and 3.94 ppm) is a marker of energy metabolism. Because creatine values are relatively stable, the ratio of another metabolite to creatine helps normalize the concentration of that metabolite.

Glutamate and glutamine represent the Glx complex (2.1 to 2.5 ppm). Glutamate is the most abundant excitatory neurotransmitter, but the Glx complex is typically best seen with very short echo times and may not be well resolved on intermediate or long echo clinical scans.

Lactate (1.33 pm) is not seen under normal conditions but is present in scenarios of anaerobic glycolysis, such as in brain ischemia and seizures. At intermediate echo times (TE = 144 msec), it is an inverted doublet, seen below the baseline, whereas at TE = 288 msec it has a positive peak.

Myoinositol (3.56 ppm) is a glial marker that is found almost exclusively in astrocytes and is believed to reflect the degree of glial proliferation.

Alanine (1.48 ppm) has a role in the citric acid cycle and can be increased in meningiomas.

Lipid (0.8 to 1.2 ppm) may be seen as a broad peak or two peaks at longer echo times, but it may be better visualized on MRS with short echo times. Lipid can be seen in areas of increased cell turnover, but its presence may also indicate voxel contamination by fat in the diploic space, the scalp, and subcutaneous tissues when the voxel is placed near one of these structures.⁹⁴

A quick assessment is frequently used to characterize the three most prominent metabolites in normal brain parenchyma: NAA, creatine, and choline. The imaginary line that connects the peaks of these metabolites normally slopes up from the left to the right at approximately 45 to 50 degrees, a so-called normal Hunter’s angle. A reduction in this angle is seen with neuronal destruction. Reversal of this angle (with the choline peak higher than the NAA peak) is strongly suggestive of tumor but can be seen with ischemic processes, inflammatory conditions, and even neurodegenerative disease.

Clinical Uses and Applications

MRS is now perhaps most widely used for the characterization of brain neoplasms (Fig. 19-26). In general, with increasing grade of glioma, one sees a reduction in the NAA peak and elevation of the choline peak (Fig. 19-27). The ultimate diagnosis is made by pathologic evaluation, but MRS can aid in determining the prognosis in some patients; for example, the percent change in the choline-NAA ratio has been shown to be useful for predicting progression of brain tumor in children. Overall, spectroscopy should be considered as a supplement to other imaging data in tumor analysis. Integration of spectroscopy data with perfusion data, enhancement pattern, and the remaining conventional MRI sequences increases the accuracy of predicting tumor grade and differentiating radiation necrosis from tumor recurrence.

FIGURE 19-26 Infiltrating medial temporal glioma. A, Axial contrast-enhanced image showing the predominant peripheral enhancement with central stippling and necrosis. The overlying grid shows voxel locations. B, A magnetic resonance spectroscopy grid as defined in A shows the greatest lipid-lactate (yellow arrowheads) concentration in the central column, whose voxels contain the greatest apparent degree of necrosis. There is also a prominent choline signal (purple arrowheads) in virtually every voxel.

FIGURE 19-27 Infiltrating glioma. A, Axial T2-weighted image demonstrating a necrotic mass centered in the medial left temporal lobe, with significant edema extending through much of the temporal lobe and into the gangliocapsular structures. B, Contrast-enhanced T1-weighted image showing heterogeneous enhancement predominantly about the periphery with finer strands of enhancement central amid necrosis. The purple box defined placement of the single spectroscopy voxel. C, Spectroscopy performed at intermediate echo time (TE = 144 msec) demonstrates an inverted lactate doublet at 1.3 to 1.4 ppm (LL), a dramatically elevated choline peak at 3.2 ppm (Cho), and a small residual N-acetylaspartate (NAA) peak at 2.0 ppm. Other metabolites, including the glial cell marker myoinositol, as well creatine (Cr) and glutamine derivatives, are also seen between 3.5 and 4.0 ppm.

In addition to tumors, recent work has shown the utility of MRS in other conditions. In seizure disorders, spectroscopy may be used to lateralize the side of hippocampal sclerosis by demonstrating decreased NAA and elevated Glx and myoinositol. It can also demonstrate bilateral temporal sclerosis, but correlation and confirmation with electroencephalography are advised.⁹⁵ MRS may detect early changes in Alzheimer’s dementia before gross findings on MRI, with decreasing NAA and increasing myoinositol observed in the occipital, temporal, parietal, and frontal regions. In imaging of patients with traumatic brain injury, reductions in the NAA-creatine and NAA-choline ratios have been shown to be a reliable index for an unfavorable outcome and may help to identify which patients have a greater likelihood of regaining consciousness.^96,97 Certain metabolic conditions have characteristic metabolite aberrations, such as the markedly elevated NAA in Canavan’s disease, which reflects a deficiency in the myelin synthesis pathway. In practice, MRS and DWI can be considered complementary methods for evaluating neurometabolic disease to give detailed information about the neurochemistry of affected brain areas.⁹⁸

Pitfalls and Limitations

The concentrations of normal metabolites in the brain vary in different age groups. Metabolites in normal subjects younger than 2 years differ from the spectral patterns seen in subjects 2 years and older (through adulthood). In these young children there is a reversal of the NAA-creatine and choline-creatine ratios. Changes are also seen in the elderly, who exhibit a normal decline in the level of NAA.

As stated earlier, the grade of an astrocytic tumor as a guide is thought to be proportional to the choline concentration. In evaluating the spectra of brain tumors, some highly malignant tumors such as glioblastoma multiforme may artifactually show low choline because of extensive necrosis. Furthermore, it is not uncommon to find some low-grade gliomas with very high choline-creatine and choline-NAA ratios. It is widely appreciated, however, that because of the heterogeneity of tumor, there is a significant overlap in the spectroscopy indices used to grade gliomas.⁹⁹