A large number of clinical and experimental studies could relate endothelial dysfunction to cardiovascular risk factors (RFs), thereby supporting the concept of endothelial dysfunction being a crucial initial step in the development of atherosclerosis (13,14 and 15). In dysfunctional endothelium, the production of NO is reduced which in turn enhances the expression of endothelial adhesion molecules, and consequently, promotes the recruitment of leucocytes to the vessel intima (16). While mononuclear leucocytes typically become foam cells by accumulating lipids and form reversible fatty streaks, accumulation of smooth muscle cells and their production of extracellular matrix may induce the formation of fibrous lesions during this phase of atherogenesis. Inflammatory mediators can impede the ability of smooth muscle cells to maintain the collagen content in atherosclerotic lesions (17). Further, activated macrophages are also known to produce matrix-degrading enzymes (18), which can cause destabilization of the fibrous cap of atheromas, finally resulting in plaque rupture (19). Repetitive plaque ruptures, often clinically silent, and subsequent healing may account for an intermittent progression of atherosclerotic lesions (20,21 and 22). With progressive disease, ruptures of larger plaques containing considerable amounts of lipids and tissue factor, a powerful procoagulant can cause ultimate thrombus formation and acute vessel occlusion (23). In line with this concept, invasive studies demonstrated an increasing risk for rupture for lesions with increasing severity of stenosis (22,24,25). Accordingly, vulnerable plaques are characterized by four major criteria (26): (1) Thin cap with large lipid core, (2) stenosis >90%, (3) endothelial denudation with superficial platelet aggregation, and (4) fissured plaque. A major goal of noninvasive imaging is to identify vulnerable plaques in the various perfusion beds, particularly the heart, the carotids, renals, and peripheral arteries of the lower limbs.

Ideally, identification of vulnerable plaques would include determination of location, mass (i.e., stenosis severity), and composition of lesions. Direct visualization of plaques in the coronary circulation has been performed in several studies and is discussed in detail in Chapters 27 and 28. In this context, it might be useful to consider the following: In a 3-mm coronary vessel imaged with a currently available spatial resolution of 0.5 × 0.5 mm² for coronary MRA, signal distribution in as little as 27 pixels would determine the degree of stenosis. With respect to plaque composition, as little as 6 pixels per component would characterize plaque tissue of a 70% stenosis (considering three components in the plaque: Fibrosis, lipid necrotic core, and inflammatory cell-rich component, and each occupying a third of plaque area) in a 3-mm vessel. Alternatively, one could aim at determining the severity of stenosis by assessing myocardial perfusion disturbances, whereby one would neglect the additional information contained in the plaque composition. In this case, a 3-mm vessel is assumed to supply approximately 60 g of muscle tissue (27), which corresponds to about 1,875 pixels (considering a spatial resolution of 2×2 mm² for perfusion imaging, slice thickness of 8 mm). It seems obvious that a change in stenosis severity would be more reliably detected by interrogating 1,875 pixels in the myocardium than 27 pixels covering the cross-sectional area of the coronary vessel. Besides these technical considerations, there is overwhelming evidence from the literature that ischemia burden is one of the most important predictors of future complications in patients with CAD (28,29 and 30). Thus, ischemia is identifying the patient population at risk which should benefit from treatment/revascularization. This concept is supported by a large number of studies demonstrating an excellent prognosis in patients without ischemia (this aspect is discussed in detail in Chapter 19). Also, the assessment of ischemia by fractional flow reserve (FFR), an invasive tool, proved the importance of ischemia as a prognostic factor and its usefulness to guide treatment decisions. In the FAME trial in >1,000 patients, treatment decisions, when based on ischemia assessment by FFR, improved patient outcome (31) and FFR-based decisions were cost-effective over coronary ang iography alone (32). In this light, the high diagnostic agreement between ischemia assessment by perfusion CMR and FFR is of particular importance (see also Table 15.2).

An elegant approach for perfusion measurements aims at exploiting the decrease in T₂ relaxation time of deoxygenated and thus, paramagnetic hemoglobin (while oxygenated hemoglobin is slightly diamagnetic causing less of T₂ shortening). With this blood oxygen level dependent (BOLD) method, a T₂– or T₂^*-weighted pulse sequence allows for an estimation of an increased content of oxygenated blood, which is expected to occur in the presence of unrestricted hyperemic flow (which is associated with a lower O₂ extraction). However, this approach is sensitive to magnetic field inhomogeneities and the robustness of the method is not yet demonstrated (51,52). Also, signal differences in normally perfused regions versus hyperemic regions (with a fourfold increase in flow) is as low as 32% in an experimental setting (53). This low signal difference might be problematic, since studies with CM first-pass approaches performed in humans showed differences in the order of 80% to 100% to be clinically inadequate while signal differences in the order of 250% to 300% were required for reliable detection of ischemic regions in patients (54). Another technique which also does not require the administration of exogenous CM is arterial spin labeling (55,56 and 57). This approach is based on T₁ measurements after global and slice-selective spin preparation (using appropriate ECG-triggered pulse sequences). Due to the inflow of unsaturated proton spins, T₁ in tissue is shortened after slice-selective preparation compared with global preparation. Assuming a two-compartment model absolute tissue perfusion is calculated. However, this concept is not yet well established for an application in humans. Magnetization transfer techniques are also under investigation to assess myocardial perfusion (58). A schematic of the various pulse sequences for perfusion assessment is given in Figure 15.1.

Most experience exists for the so-called CM first-pass approaches used for the detection of perfusion abnormalities. With this technique, a bolus of CM is injected into a peripheral vein and its effect on myocardial signal during hyperemia (stress-only protocol) and during resting conditions (stress-rest protocol) is monitored with fast MR pulse sequences. Common for all CM first-pass techniques either under development or in clinical routine are the requirements to (1) provide high spatial resolution to permit detection of small subendocardial perfusion deficits, (2) provide adequate cardiac coverage to allow for assessment of the extent of perfusion deficits, and (3) feature high CM sensitivity to generate optimum contrast between normally and abnormally perfused myocardium during CM first pass. Finally, acquisition of perfusion data should occur every 1 to 2 heartbeats to yield signal intensity—time curves of adequate temporal resolution which allow for extraction of various perfusion parameters (see below). To fulfill these requirements, speed of data acquisition and time-efficient magnetization preparation are crucial. Since passage of CM during hyperemic conditions only last 5 to 10 seconds, breathing motion is typically minimized by a breath-hold maneuver while cardiac motion is eliminated by ECG triggering. In contrast to scintigraphic techniques, this control of cardiac and breathing motion preserves the high spatial resolution of the MR perfusion data in the order of 1 to 3 × 1 to 3 mm. Most commonly used are the extravascular gadolinium chelates in combination with heavily T₁-weighted pulse sequences. Signal increase in the myocardium during first-pass of these CM is dependent on both, perfusion of the myocardium and diffusion of CM into the interstitial space. Since these CM are excluded from the intracellular compartment, that is, of viable cells with intact cell membranes, a perfusion deficit during first pass can reflect either hypoperfused viable myocardium or scar tissue. Conversely, for scintigraphic techniques, the flow tracers are taken up by viable myocytes causing tracer accumulation, and consequently, an increase in signal (a schematic is given in Figure 15.2). In order to differentiate by MR imaging viable hypoperfused myocardium from scar, both demonstrating slow CM wash-in during MR first-pass studies, it is recommended to wait for establishment of equilibrium distribution of CM in the various compartments, that is, blood and extracellular, interstitial space (of viable myocardium and scar), which occurs within 10 to 20 minutes post-CM injection. During this condition, late enhancement imaging (with the inversion time set to null normal myocardium) displays hypoperfused but viable myocardium as dark tissue, while scar tissue appears bright (due to the high distribution volume of CM in the enlarged interstitial space of fibrous tissue). For comparison, during equilibrium distribution of scintigraphic tracers (occurring after
rest injection or rest re-injection) tracer is not taken up by scar tissue, which consequently appears as a cold spot, while viable tissue appears as a hot spot (see also Figure 15.2).

MAGNETIZATION PREPARATION. In the beginning of firstpass perfusion-CMR imaging, a high contrast between abnormally and normally perfused myocardium during peak effect of bolus was achieved by preparing the myocardium by a 180-degree inversion pulse (IR technique). To obtain maximum contrast, the signal of precontrast myocardium was nulled by applying inversion recovery times in the order of 300 to 400 milliseconds (time from preparation to recovery, i.e., readout of central k-space lines: TPR) (59,60,61,62,63,64,65,66 and 67). However, combining this IR-preparation scheme with a turboFLASH (fast low-angle shot), readout resulted in long acquisition windows of 650 to 750 milliseconds per slice, which precludes the acquisition of multiple slices during 1 to 2 heartbeats. Partial preparation flip angles, for example, of 45 to 60 degrees were suggested which allow for shorter delay times (TPR) (68,69). However, this approach limited the dynamic range of signal response (54). Therefore, nowadays a saturation recovery preparation by a 90-degree pulse is typically applied, since it shortens the delay time TPR to 100 to 150 milliseconds (54), and additionally it renders the sequence heart rate independent. Simulations in Figure 15.3 demonstrate the influence of increasing TPR following a 90-degree preparation on signal response for a fast gradientecho pulse sequence (TR/TE: 5/1 milliseconds, r₁ and r₂^* of CM is 4 and 6/mmol/sec, respectively). With increasing TPR the sensitivity of the pulse sequence for low CM concentrations increases, however, at the cost of reduced sensitivity of higher CM concentrations. In order to select the optimum TPR, not only the signal intensity—CM concentration relationship is important—but also the number of excitations required achieving steady state of magnetization. Further, TPR should result in acquisition windows which are placed within phases of minimal motion during the cardiac cycle. A TPR of 120 milliseconds in combination with a hybrid echoplanar readout scheme can result in an acquisition window as short as 240 milliseconds per slice (33). This combination avoids data acquisitions during rapid cardiac motion, that is, during early systole, early diastole, and late diastole (atrial contraction) over a wide range of heart rates which may occur during hyperemia. This very fast approach allows for multislice data acquisition, while preserving a high sampling rate. To speed up the acquisition, it was suggested to play out one single nonslice-selective 90-degree saturation pulse, and to acquire all slices in series thereafter (36). With this scheme, however, the delay time varies from slice to slice, and consequently, CM sensitivity becomes dependent upon
slice position, which potentially complicates data analysis. In a modified version of a saturation preparation, the entire myocardium is experiencing a 90-degree saturation pulse except that slice of tissue which is immediately imaged after preparation (70). With this scheme the time period for readout of slice_n is utilized to prepare slice_n+1 and so on, that is, the acquisition window equals the saturation recovery time. This approach allows acquisitions of up to seven slices every 2 heartbeats at heart rates below 115 beats per minute.

MR DATA READOUT. Conventional TurboFlash pulse sequences were often used in the past for perfusion imaging. With this technique each k-line is preceded by a radiofrequency excitation, which results in a readout duration in the order of 350 to 450 milliseconds depending on the number of phase-encoding steps and duration of TR (59,60,61,62,63,64,65 and 66). Since this acquisition window is too long to allow for multislice imaging, faster echo-planar pulse sequences became popular (71,72 and 73), particularly when broken up in fewer lines per radiofrequency excitation, known as hybrid echo-planar pulse sequences (33,54,68,69). This technique reduces TE, and consequently, renders the sequence more robust with respect to potential susceptibility artifacts. With these accelerated pulse sequences several k-lines are acquired following one single radiofrequency excitation reducing the TR per k-line down to <2 milliseconds and consequently, reducing total acquisition window/slice substantially. Since fast imaging is inherently conflicting with lower signal-to-noise ratio, readout strategies during steady-state conditions of magnetization appear promising, since they preserve magnetization and thus, a high signal-to-noise ratio. This technique has been employed successfully in animal models for monitoring CM first pass through the myocardium (74). In a human volunteer study, however, ECG triggering was problematic as well as the presence of artifacts (banding artifact probably due to off-resonance) and a low signal increase (approximately 40% of baseline signal) (75). A high image quality, reflected by a substantial signal increase in the myocardium during first pass, appears crucial for reliable CAD detection, and even offsets some compromise in cardiac coverage (54). Nevertheless, improving cardiac coverage without the need to reduce spatial and temporal resolution would be beneficial, since extent of CAD correlates with outcome. Parallel imaging approaches can increase coverage without compromise in spatial and temporal resolution. In a recent study in human volunteers, the loss in signal-to-noise ratio given by g × vR (with g being the so-called geometry factor and R the accelerating factor) was compensated for by a longer TR (due to implementation of TSENSE, a modification of SENSE (76)) and increasing the readout flip angle from 20 to 30 degrees. This accelerated hybrid echo-planar saturation recovery technique yielded twice as many slices as obtained with the nonparallel approach, while signal-tonoise ratio improved by approximately 20% (77). Finally, even faster imaging is possible by exploiting correlations in k-space and time (k-t) combined with sensitivity encoding (78) yielding spatial resolutions of 1.4 × 1.4 mm² with high diagnostic yield (79,80). This approach was recently used to acquire three-dimensional (3D) myocardial perfusion data at 3 T with a spatial resolution of 2.3 × 2.3 × 10 mm³ (16 slices) acquired within a 200-millisecond window (40). While a 3D acquisition might be advantageous with respect to SNR and registration during postprocessing, larger studies, ideally, multicenter trials would be needed to show an incremental benefit of these highly accelerated strategies versus conventional perfusion sequences.

IMAGING PARAMETERS AND MYOCARDIAL SIGNAL RESPONSE. The concept of CM first-pass perfusion imaging is based on the assumption that the CM administered intravenously is delivered to the myocardium in relation to tissue perfusion, and consequently, the relationship between signal intensity and CM concentration should be known. With fast gradient-echo pulse sequences the extravascular Gd-chelates typically show an increase in myocardial signal intensity up to a maximum (T₁-dominated behavior), and signal decreases at higher CM concentrations (T₂-dominated behavior). It is desirable that the pulse sequence applied shows a linear CM concentration — signal intensity relationship. This relationship is dependent on the type of pulse sequence and the imaging parameters such as magnetization preparation, delay time between preparation and readout (TPR), readout flip angle, k-space trajectories during readout, and others. The simulations in Figure 15.3 demonstrate that signal response is not only influenced by TPR, but also be the readout flip angle. During the readout period, all the curves shown in Figures 15.3A-C approach steadystate magnetization which yields the curves shown in Figure 15.3D. Varying signal responses for different TR periods, however, cause signal inhomogeneities in k-space, and thus, image artifacts. The simulations in Figure 15.4 show the course of signal response during successive TR periods in relation to TPR and readout flip angle. At higher flip angles (e.g., 55 degrees), the steady state is approached earlier (= 6 TR periods at a TPR of 130 milliseconds), while a shorter TPR of 11.6 milliseconds achieves steady state already after first TR. These considerations may illustrate that imaging parameters strongly affect signal response and should be optimized with respect to dynamic range of signal response, signal intensity-CM concentration relationship, homogeneity of signal in k-space, and timing of acquisition during the cardiac cycle. In a study in phantoms, it could be shown that optimization of a hybrid echo-planar pulse sequence (68) with respect to TPR, readout flip angle and preparation flip angle, improved signal response of Gd-DTPA-doped phantoms from 80% (of nondoped phantom) to 250%, which ultimately increased diagnostic performance in patients significantly (54). However, changes in these parameters typically go along with changes in the acquisition window/slice which can affect cardiac coverage.

Ideally, perfusion data would be evaluated by dedicated algorithms rather than by subjective reading (see below). This approach would allow comparing signal response in an individual patient with a normal database, which could render the technique less observer dependent. Such observer-independent comparisons with a normal database would require adaptation of normal values in case, the pulse sequence and/or imaging parameters had been changed. These technical considerations illustrate that myocardial signal response is strongly dependent on imaging parameters. Therefore, two important multicenter prospective trials (MR-IMPACT I and II) proved the high
diagnostic yield of pulse sequences being based on comparable sequence parameters implemented on various vendor platforms (12,44).

These CM are most frequently used and represent the conventional extravascular Gd-chelates. Administration of these CM typically induce an increase in myocardial signal during first pass, which allows detection of hypoperfused myocardium as territories of lower signal by visual assessment, or by a quantitative approach extracting various parameters from the signal intensity-time curves (see below). Studies demonstrated a strong impact of maximum signal increase
in the myocardium achievable during first pass and diagnostic performance (54). Therefore, best test performance is expected for high CM doses (in combination with pulse sequences with strong T₁ weighting), which provide a large dynamic range of signal response in the myocardium, that is, a large signal difference between normally perfused and hypoperfused territories. However, at higher CM doses, susceptibility artifacts may occur at the subendocardium, where differences in CM concentrations between blood and myocardium are particularly high during first pass. To optimize the CM dose (providing high signal increase in the myocardium without causing susceptibility artifacts in the subendocardium), several multicenter MR perfusion trials were performed. For strongly T₁-weighted sequences, they showed absence of a susceptibility-induced signal drop in the subendocardial layer up to doses of 0.15 mmol/kg of a Gd-chelate (9).

The safety of Gd-based CM when applied for stress perfusion CMR was recently reported from the European CMR registry (83). In 17,767 patients, no severe adverse event occurred. Thirty acute adverse reactions were observed (0.17%), all classified as mild according to the American College of Radiology definition (83).

T₁-enhancing intravascular CM such as albumin-targeted MS-325 (69) or poly-lysine-Gd (84,85) have been applied in animal models. In these studies, differences (for group means) between normally perfused and stenosis-dependent, hypoperfused myocardium were reported. However, as to our knowledge, these intravascular Gd-based CM have not yet been tested in humans. Intravascular superparamagnetic iron oxide nanoparticles (USPIO) with a starch coating were used for perfusion studies in humans (86). In combination with a T₂-weighted turbo spin-echo sequence, a signal drop in normal myocardium of 59% (in eight volunteers) was observed (this CM is no longer under investigation due to iron accumulation in the liver). Since susceptibility influences signal drop of T₂^*-enhancing CM, not only the concentration of CM in the voxel but also intravoxel distribution (homogeneous vs. inhomogeneous) determines its T₂^*-shortening effect, thus, rendering such an approach susceptible to vessel architecture (orientation, inter-vessel distance) within the myocardium. This principle was demonstrated with a dysprosium chelate in an elegant experimental model of reperfusion injury (87). Since T₂^* approaches are affected by these geometry factors (which are not known a priori), measurements of CM concentrations by T₁ techniques are considered superior.

Conventional Gd-chelates act by modulating the signal from surrounding water molecules by accelerating their relaxation rates. The low polarization level of the spin population of the NMR-active nuclei at the thermal equilibrium at a field strength of 1.5 T is compensated by the vast abundance of water molecules. One way to increase signal could be achieved by increasing the field strength, since the polarization level increases with magnetic flux density. Alternatively, however, the polarization level of the spin population of specific nuclei, such as liquid ¹³C in various compounds, can be increased by a factor of up to 100,000 compared with polarization of water protons at the thermal equilibrium by two techniques, that is, dynamic nuclear polarization (88) and parahydrogen-induced polarization (89). A major advantage of hyperpolarized ¹³C-CM is the fact that signal is only received from the ¹³C-nuclei, thus, no signal from background tissue is obtained. This theoretically allows for absolute quantification of perfusion, since signal is directly proportional to the amount of ¹³C-molecules (90). However, since the spin population of ¹³C-molecules is far away from thermal equilibrium, application of RF pulsing irreversibly destroys longitudinal magnetization. This means that CM depolarization must be taken into account similarly to radiotracer modeling which corrects for tracer decay. In addition, depolarization due to RF pulsing has to be considered, which is depending on the pulse sequence and the imaging parameters. Johansson et al. (91) showed that depolarization can be approximated by a monoexponential function with a time constant T_D and successfully applied this concept for cerebral perfusion quantification.