DOI: 10.1148/radiol.2412041866
(Radiology 2006;241:338-354.)
© RSNA, 2006
Cardiac MR Imaging: State of the Technology1
J. Paul Finn, MD,
Kambiz Nael, MD,
Vibhas Deshpande, PhD,
Osman Ratib, MD, PhD and
Gerhard Laub, PhD
1 From the Department of Radiological Sciences, David Geffen School of Medicine, University of California Los Angeles, 10945 Le Conte Ave, Suite 3371, Los Angeles, CA 90095-7206 (J.P.F., K.N., O.R.), and Siemens Medical Solutions, Los Angeles, Calif (V.D., G.L.). Received November 2, 2004; revision requested January 3, 2005; revision received June 24; accepted July 20; final version accepted November 23; final review by J.P.F. May 16, 2006.
Address correspondence to J.P.F. (e-mail: pfinn{at}mednet.ucla.edu).
 |
ABSTRACT
|
|---|
Recent developments in magnetic resonance (MR) imaging of the heart have refocused attention on the potential of MR and continue to attract intense interest within the radiology and cardiology communities. Improvements in speed, image quality, reliability, and range of applications have evolved to the point where cardiac MR imaging is increasingly seen as a practical clinical tool. As is often the case with MR imaging, not all of the most powerful techniques are necessarily easy to master or understand, and many—nonspecialists and specialists alike—are challenged to stay abreast. This review covers some of the major milestones that have led to the current state of cardiac MR and attempts to put into context some concepts that, although technical, have a real impact on the diagnostic power of cardiac MR imaging. Topics discussed include functional imaging, myocardial viability and perfusion imaging, flow quantification, and coronary artery imaging. A review such as this can only scratch the surface of what is a dynamic interdisciplinary field, but the hope is that sufficient information and insight are provided to stimulate the motivated reader to take his or her interest to the next level.
© RSNA, 2006
|
INTRODUCTION
|
|---|
Although the potential of magnetic resonance (MR) imaging as a tool for noninvasive diagnosis of heart disease has long been recognized, its widespread adoption in clinical practice is only recently gaining pace. At various times, MR imaging has been hailed as the single modality capable of defining cardiac anatomy and function, myocardial perfusion, myocardial viability, and coronary artery anatomy (1,2). However, cardiac MR imaging has often been regarded as a difficult test with unpredictable results. In recent years, technical developments have had a dramatic effect on cardiac MR applications, such that the debate no longer focuses on the diagnostic power of MR imaging but on availability or local expertise.
More than any other organ system, the heart has played host to a perplexing evolution of MR techniques. Some of these methods, although highly sophisticated, have limited clinical applications. Others have found their way into clinical practice or stand poised to do so. In all cases, MR imaging of the heart must deal with cardiac and respiratory motion, with the conflicting requirements for high spatial and temporal resolution, and with the demand for accurate and reproducible measurements in a clinical environment.
Recognizing that development is an ongoing process, in this review we emphasize those techniques for which there are established clinical applications or that we believe will have a clinical role in the near term. We believe that important tools for today's cardiac MR imager include (a) fast, high-spatial-resolution, steady-state techniques for cine imaging and coronary angiography; (b) dependable techniques for T1-weighted imaging of contrast material–avid myocardial scar; (c) fast techniques to capture the first pass of intravenous contrast material through the heart; and (d) the option to apply parallel imaging to all of the above. The fact that few of these tools were widely available 4 years ago reflects well on the pace of development of the field and the speed of adoption by the community.
|
BLACK-BLOOD ANATOMIC IMAGING
|
|---|
A necessary first step in cardiac MR imaging was the application of gating to spin-echo imaging (3–5). The resulting "dark blood" images of the myocardium and cardiac chambers focused attention on the outstanding potential of MR imaging of the heart (6–11). Today, spin-echo imaging plays a more secondary role, but for specific applications involving structural abnormalities of the ventricles and the pericardium, it may still prove useful (12,13). Traditional spin-echo imaging has given way to echo-train imaging, which is usually performed with breath holding.
Black-blood preparation schemes are now standard for spin-echo imaging of the heart and blood vessels, and this usually involves a double inversion pulse pair (Fig 1a) (14–16). With this approach, a single section is acquired per breath hold, and a full examination can be time consuming. At the extreme end of echo-train imaging is the single-shot method (half-Fourier rapid acquisition with relaxation enhancement), with which an image can be acquired within a single heartbeat (17,18) (Fig 1b, 1c). For many protocols involving anatomic surveys, single-shot spin-echo imaging can be a valuable supplement.
View larger version (21K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1a: (a) Diagram of black-blood double-inversion turbo spin-echo MR sequence. When the electrocardiographic (ECG) R wave is detected, a spatially nonselective (non-sel) inversion pulse is applied and immediately followed by section-selective (sel) reversion pulse. The net effect is to leave spins within the section unaffected, while spins outside the section (including flowing blood) are inverted. Over time, inverted spins relax toward zero where, even if excited by subsequent radiofrequency pulses, they generate little signal. If the readout module, a 90° pulse followed by a train of refocusing 180° pulses, is applied when blood is relaxing to zero, inflowing blood produces no signal. TI = inversion time. (b) Diagram of black-blood double-inversion single-shot spin-echo echo-train sequence. Black-blood preparation is identical to that in a. In this case, however, all lines for the complete image are read out in a single heartbeat (ie, data acquisition is not "segmented"). (c) Adenocarcinoma of right lung invading left atrium. Coronal black-blood double-inversion single shot spin-echo echo-train MR image (repetition time (TR) msec/echo time msec, 2000/56; flip angle, 90°) was acquired in a single heartbeat. Note relatively uniform low signal intensity from blood in cardiac chambers and major vessels and how well the tumor (arrow) is shown invading left atrium from the right lung.
|
|
View larger version (15K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1b: (a) Diagram of black-blood double-inversion turbo spin-echo MR sequence. When the electrocardiographic (ECG) R wave is detected, a spatially nonselective (non-sel) inversion pulse is applied and immediately followed by section-selective (sel) reversion pulse. The net effect is to leave spins within the section unaffected, while spins outside the section (including flowing blood) are inverted. Over time, inverted spins relax toward zero where, even if excited by subsequent radiofrequency pulses, they generate little signal. If the readout module, a 90° pulse followed by a train of refocusing 180° pulses, is applied when blood is relaxing to zero, inflowing blood produces no signal. TI = inversion time. (b) Diagram of black-blood double-inversion single-shot spin-echo echo-train sequence. Black-blood preparation is identical to that in a. In this case, however, all lines for the complete image are read out in a single heartbeat (ie, data acquisition is not "segmented"). (c) Adenocarcinoma of right lung invading left atrium. Coronal black-blood double-inversion single shot spin-echo echo-train MR image (repetition time (TR) msec/echo time msec, 2000/56; flip angle, 90°) was acquired in a single heartbeat. Note relatively uniform low signal intensity from blood in cardiac chambers and major vessels and how well the tumor (arrow) is shown invading left atrium from the right lung.
|
|
View larger version (118K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1c: (a) Diagram of black-blood double-inversion turbo spin-echo MR sequence. When the electrocardiographic (ECG) R wave is detected, a spatially nonselective (non-sel) inversion pulse is applied and immediately followed by section-selective (sel) reversion pulse. The net effect is to leave spins within the section unaffected, while spins outside the section (including flowing blood) are inverted. Over time, inverted spins relax toward zero where, even if excited by subsequent radiofrequency pulses, they generate little signal. If the readout module, a 90° pulse followed by a train of refocusing 180° pulses, is applied when blood is relaxing to zero, inflowing blood produces no signal. TI = inversion time. (b) Diagram of black-blood double-inversion single-shot spin-echo echo-train sequence. Black-blood preparation is identical to that in a. In this case, however, all lines for the complete image are read out in a single heartbeat (ie, data acquisition is not "segmented"). (c) Adenocarcinoma of right lung invading left atrium. Coronal black-blood double-inversion single shot spin-echo echo-train MR image (repetition time (TR) msec/echo time msec, 2000/56; flip angle, 90°) was acquired in a single heartbeat. Note relatively uniform low signal intensity from blood in cardiac chambers and major vessels and how well the tumor (arrow) is shown invading left atrium from the right lung.
|
|
|
CARDIAC CINE IMAGING
|
|---|
Cine imaging of the beating heart remains the cornerstone of functional cardiac MR imaging. The early work in MR imaging of the heart involved the use of spoiled gradient-echo (GRE) pulse sequences to generate gated cine (19,20) and gated velocity imaging (20–23). A single phase-encoding view of multiple time points in the cardiac cycle was acquired in each heartbeat, gated to the ECG trace, so the time required, for example, for 128 phase-encoding steps was at least 128 cardiac cycles, or about 2 minutes (Fig 2a). In practice, breathing motion artifact was managed by averaging at least two excitations, so the actual acquisition time per section was closer to 4 minutes.
View larger version (17K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2a: Diagrams of (a–c) prospectively ECG-triggered and (d) retrospectively ECG-gated cine MR pulse sequences. Ts = imaging time. (a) After the R wave, the first line of the first cardiac phase is acquired, followed in sequence by the first line of all the other cardiac phases, up to the final phase acquired. In the next cardiac cycle, the second line of all the cardiac phases is acquired, up to Np. This continues until all lines (N, phase-encoding steps) have been acquired, or for N heartbeats. In this scheme, the acquisition window is prescribed ahead of time and is generally chosen to be less than the average R-R interval so that the next R wave is not missed, should it occur earlier than anticipated. For this reason, the last 10% of diastole is usually not sampled with prospective triggering. (b) k-Space segmentation. After the R wave, first five lines (in this example) of first cardiac phase are acquired, followed in sequence by first five lines of all the other cardiac phases, up to Np. In the next cardiac cycle, next five lines of all cardiac phases are acquired, up to Np. This continues until all lines (phase-encoding steps) have been acquired, or for N/5 heartbeats. In general, if there are Ls lines per segment, it will take N/Ls heartbeats to acquire the cine study, and the duration of each cardiac phase will be Ls times the duration of the nonsegmented sequence. In this way, temporal resolution is traded against acquisition time. (c) k-Space segmentation and echo sharing. Similar to b, but central line of each segment is repeated between segments, and data are shared around this additional line. In this way, a "sliding temporal window" is generated that is five lines wide but is updated almost twice as often as in the non–echo-shared version. Therefore, almost twice as many cardiac phases are generated. (d) Similar to a, but acquisition window extends to the next R wave, sampling the entire cardiac cycle. Data are then re-sorted according to their location relative to the R-R interval.
|
|
An important advance was the introduction of breath-hold cine MR imaging with k-space segmentation (15). The concept of segmenting k-space in this way represented a huge step forward and was followed by the first descriptions of breath-hold coronary MR imaging and coronary flow quantification by the same group (24). Breath-hold cine imaging involves acquisition of multiple lines (phase-encoding steps) of data after a single ECG trigger pulse, the sum of these lines forming a segment of k-space. Relative to conventional (nonsegmented) data acquisition, the new method accelerated image capture by the number of lines in the segment—typically about seven (Fig 2b). So, what used to take 140 heartbeats to acquire could now be acquired in 20, albeit with a trade-off in temporal resolution.
View larger version (19K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2b: Diagrams of (a–c) prospectively ECG-triggered and (d) retrospectively ECG-gated cine MR pulse sequences. Ts = imaging time. (a) After the R wave, the first line of the first cardiac phase is acquired, followed in sequence by the first line of all the other cardiac phases, up to the final phase acquired. In the next cardiac cycle, the second line of all the cardiac phases is acquired, up to Np. This continues until all lines (N, phase-encoding steps) have been acquired, or for N heartbeats. In this scheme, the acquisition window is prescribed ahead of time and is generally chosen to be less than the average R-R interval so that the next R wave is not missed, should it occur earlier than anticipated. For this reason, the last 10% of diastole is usually not sampled with prospective triggering. (b) k-Space segmentation. After the R wave, first five lines (in this example) of first cardiac phase are acquired, followed in sequence by first five lines of all the other cardiac phases, up to Np. In the next cardiac cycle, next five lines of all cardiac phases are acquired, up to Np. This continues until all lines (phase-encoding steps) have been acquired, or for N/5 heartbeats. In general, if there are Ls lines per segment, it will take N/Ls heartbeats to acquire the cine study, and the duration of each cardiac phase will be Ls times the duration of the nonsegmented sequence. In this way, temporal resolution is traded against acquisition time. (c) k-Space segmentation and echo sharing. Similar to b, but central line of each segment is repeated between segments, and data are shared around this additional line. In this way, a "sliding temporal window" is generated that is five lines wide but is updated almost twice as often as in the non–echo-shared version. Therefore, almost twice as many cardiac phases are generated. (d) Similar to a, but acquisition window extends to the next R wave, sampling the entire cardiac cycle. Data are then re-sorted according to their location relative to the R-R interval.
|
|
The implementation of echo sharing improved temporal resolution substantially while keeping the acquisition time to a manageable breath hold. Echo-sharing involves acquisition of the central k-space point twice as often as the other k-space points and selection of appropriate neighboring points to fill the k-space segment. In this way, the number of frames (temporal resolution) is doubled with little increase in imaging time (Fig 2c).
View larger version (14K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2c: Diagrams of (a–c) prospectively ECG-triggered and (d) retrospectively ECG-gated cine MR pulse sequences. Ts = imaging time. (a) After the R wave, the first line of the first cardiac phase is acquired, followed in sequence by the first line of all the other cardiac phases, up to the final phase acquired. In the next cardiac cycle, the second line of all the cardiac phases is acquired, up to Np. This continues until all lines (N, phase-encoding steps) have been acquired, or for N heartbeats. In this scheme, the acquisition window is prescribed ahead of time and is generally chosen to be less than the average R-R interval so that the next R wave is not missed, should it occur earlier than anticipated. For this reason, the last 10% of diastole is usually not sampled with prospective triggering. (b) k-Space segmentation. After the R wave, first five lines (in this example) of first cardiac phase are acquired, followed in sequence by first five lines of all the other cardiac phases, up to Np. In the next cardiac cycle, next five lines of all cardiac phases are acquired, up to Np. This continues until all lines (phase-encoding steps) have been acquired, or for N/5 heartbeats. In general, if there are Ls lines per segment, it will take N/Ls heartbeats to acquire the cine study, and the duration of each cardiac phase will be Ls times the duration of the nonsegmented sequence. In this way, temporal resolution is traded against acquisition time. (c) k-Space segmentation and echo sharing. Similar to b, but central line of each segment is repeated between segments, and data are shared around this additional line. In this way, a "sliding temporal window" is generated that is five lines wide but is updated almost twice as often as in the non–echo-shared version. Therefore, almost twice as many cardiac phases are generated. (d) Similar to a, but acquisition window extends to the next R wave, sampling the entire cardiac cycle. Data are then re-sorted according to their location relative to the R-R interval.
|
|
In its most recent implementation, cardiac triggering can now be combined with retrospective gating to maximize time efficiency while the entire cardiac cycle is sampled (Fig 2d). Because the entire heart can be sampled with MR imaging (without limitations in acoustic windows, as sometimes occurs with echocardiography), it is feasible to measure ventricular dimensions and myocardial mass with high accuracy and reproducibility (25–28). Methods are available for manual or semiautomated segmentation of the myocardial borders, allowing derivation of chamber volumes, ejection fraction, cardiac output, and ventricular muscle mass (29).
View larger version (14K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2d: Diagrams of (a–c) prospectively ECG-triggered and (d) retrospectively ECG-gated cine MR pulse sequences. Ts = imaging time. (a) After the R wave, the first line of the first cardiac phase is acquired, followed in sequence by the first line of all the other cardiac phases, up to the final phase acquired. In the next cardiac cycle, the second line of all the cardiac phases is acquired, up to Np. This continues until all lines (N, phase-encoding steps) have been acquired, or for N heartbeats. In this scheme, the acquisition window is prescribed ahead of time and is generally chosen to be less than the average R-R interval so that the next R wave is not missed, should it occur earlier than anticipated. For this reason, the last 10% of diastole is usually not sampled with prospective triggering. (b) k-Space segmentation. After the R wave, first five lines (in this example) of first cardiac phase are acquired, followed in sequence by first five lines of all the other cardiac phases, up to Np. In the next cardiac cycle, next five lines of all cardiac phases are acquired, up to Np. This continues until all lines (phase-encoding steps) have been acquired, or for N/5 heartbeats. In general, if there are Ls lines per segment, it will take N/Ls heartbeats to acquire the cine study, and the duration of each cardiac phase will be Ls times the duration of the nonsegmented sequence. In this way, temporal resolution is traded against acquisition time. (c) k-Space segmentation and echo sharing. Similar to b, but central line of each segment is repeated between segments, and data are shared around this additional line. In this way, a "sliding temporal window" is generated that is five lines wide but is updated almost twice as often as in the non–echo-shared version. Therefore, almost twice as many cardiac phases are generated. (d) Similar to a, but acquisition window extends to the next R wave, sampling the entire cardiac cycle. Data are then re-sorted according to their location relative to the R-R interval.
|
|
Breath-hold cine MR imaging with a spoiled GRE sequence remained the standard for functional imaging of the heart until the introduction of steady-state techniques at the turn of the century.
|
CINE MR WITH STEADY-STATE FREE PRECESSION
|
|---|
In both its original and segmented form, spoiled GRE cine MR is largely a T1-weighted sequence. As such, it is dependent on through-plane flow enhancement to generate contrast between the blood and the myocardium, analogous to time-of-flight MR angiography (30). Like time-of-flight imaging, if the TR is too short or the flow is too slow, the blood becomes saturated. This is particularly the case for long-axis imaging, where blood may linger in the section, and even for short-axis imaging if myocardial function is poor. So, although hardware technology had evolved to the point where a very short TR was achievable, the dependency on flow-enhancement severely restricted the minimum TR that could be used in practice.
An effective alternative to the spoiled GRE approach is steady-state free precession (SSFP) (31,32). Originally described in the heart for real-time cine MR imaging with a very short TR, it proved possible to maintain the steady state for gated high-spatial-resolution cine imaging with k-space segmentation (33,34). In this form, segmented k-space SSFP cine MR imaging has largely replaced spoiled GRE cine MR imaging for all routine applications (35).
Among the advantages of SSFP cine imaging are the relative independence of contrast from blood flow, the speed of acquisition, and the high contrast-to-noise ratio per unit time. With SSFP, the blood signal is dependent on its relaxation time (specifically, the T2*/T1 ratio) rather than its speed of flow, so that even with very poor systolic function, blood-myocardial contrast is excellent. This property also makes myocardial segmentation more easily automated (36) and enhances the accuracy of myocardial mass measurements (37). SSFP techniques recycle coherent transverse magnetization by completely rewinding all gradients and alternating the phase of the radiofrequency excitation (31) (Fig 3a). SSFP cine works best with the shortest possible TR (on the order of 3 msec or faster) (32), so imaging speed is almost three times greater than that of spoiled GRE cine MR (34). A cine SSFP set with a high contrast-to-noise ratio can be acquired in as little as 7 seconds. Also, although it seems counterintuitive, SSFP cine imaging yields highest contrast-to-noise ratio when the largest flip angle achievable is used (Fig 3b). In practice, at a TR of 2.5–3.5 msec, the allowed specific absorption rate will generally limit the flip angle to 60°–70° at 1.5 T.
View larger version (21K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3a: (a) Diagram of SSFP pulse sequence. Gradients (G) are fully balanced along all three (section-selective, phase-encoding, and readout) axes. Between each two radiofrequency pulses, the sum of positive gradient areas is exactly balanced by the sum of negative gradient areas. In this case, echo and readout occur midway between radiofrequency pulses. TE = echo time. (b) Short-axis breath-hold cine MR images. Comparison of SSFP (left column: 3/1.5; flip angle, 60°) and spoiled GRE (right column: 8/4; flip angle, 20°) cine images. Pericardial fluid (arrows, top row) has higher signal intensity on SSFP image (top left). Blood signal and myocardial definition (arrows, bottom row) are also better on SSFP image (bottom left). Arrowheads = papillary muscle. (Adapted and reprinted, with permission, from reference 34.)
|
|
View larger version (162K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3b: (a) Diagram of SSFP pulse sequence. Gradients (G) are fully balanced along all three (section-selective, phase-encoding, and readout) axes. Between each two radiofrequency pulses, the sum of positive gradient areas is exactly balanced by the sum of negative gradient areas. In this case, echo and readout occur midway between radiofrequency pulses. TE = echo time. (b) Short-axis breath-hold cine MR images. Comparison of SSFP (left column: 3/1.5; flip angle, 60°) and spoiled GRE (right column: 8/4; flip angle, 20°) cine images. Pericardial fluid (arrows, top row) has higher signal intensity on SSFP image (top left). Blood signal and myocardial definition (arrows, bottom row) are also better on SSFP image (bottom left). Arrowheads = papillary muscle. (Adapted and reprinted, with permission, from reference 34.)
|
|
The only real drawback of SSFP cine is its sensitivity to off-resonance effects. These are manifest as dark bands in the image that, depending on where they occur, may be unimportant or cause troublesome artifact. If a band happens to cross a major inflow vessel or a cardiac chamber, it disrupts the steady state and can degrade contrast and cause artifact. The distance between these bands is inversely related to the TR, which is why the shortest possible TR is desirable. The homogeneity of the main magnetic field is a major factor in determining the location of the dark bands on SSFP images (38). If the field is very uniform, the bands are widely spaced. If, on the other hand, the field is nonuniform, steep frequency gradients cause the bands to be closer together and to more likely cause artifact by disrupting the steady state. The recipe for high-quality SSFP cine is, therefore, a very short TR, a large flip angle, and a very uniform magnetic field. In practice, the patient's own magnetic susceptibility sometimes modifies the static field in ways that are difficult to counteract.
With the recent introduction of improved radiofrequency receiver chains and parallel imaging techniques (39,40), the performance of SSFP cine MR has been further enhanced (41). Parallel imaging technology makes use of the spatial sensitivity profiles of surface coils to provide information about where the MR signal is coming from. By combining such information with the spatial encoding of the imaging gradients, the number of gradient encodings (phase-encoding steps) necessary to reconstruct an image can be reduced several fold. For example, by halving the number of gradient encodings (by skipping every second one), the field of view in the phase-encoding direction is halved and the acquisition is accelerated by a factor of two. Normally, this would result in an overfolding artifact, but parallel acquisition techniques prevent or "unwrap" the overfolding by using the information from the surface coils.
Two broad approaches to parallel imaging are in general use. The methods described by Sodickson and Manning (39) and the modifications described by Griswold et al (42) work by substituting directly for phase-encoding steps prior to Fourier transformation. The unmeasured intermediate k-space lines are derived from the measured ones by multiplying each measured signal by a sinusoid, which is approximated by an array of surface coils. Such approaches are said to work in the k-space domain, and they prevent overfolding. The second approach, originally described by Pruessmann et al (43), works on the image data after Fourier transformation. Here, the sensitivity profile of an array of surface coils is used to unwrap the overfolding by means of the mathematic process of matrix inversion.
Parallel acquisition is a general tool with applications across a wide spectrum of cardiac and noncardiac techniques. When used appropriately and with suitable radiofrequency coils and channels, it can enhance speed, coverage, or both (43). Depending on the quality of the information available from the surface coils, acquisition speed may be enhanced several fold.
A drawback of parallel acquisition is the decrease in signal-to-noise ratio (SNR), which invariably occurs. Several factors may contribute to this decrease in SNR, including a geometry factor determined by the coils, a decrease in acquisition time (if acquisition speed is increased), and a decrease in voxel size (if image resolution is increased). Because it is intrinsically a high-SNR technique, SSFP cine MR is more tolerant of the SNR penalty inherent in parallel imaging than is spoiled GRE cine MR. For parallel imaging with an acceleration factor of two, a diagnostic-quality cine set can be acquired in about 4 seconds, or multiple sections can be acquired in a single reasonable breath hold.
It should be noted, however, that whereas parallel imaging can speed image acquisition and help minimize susceptibility artifacts with multiecho techniques such as echo-planar imaging, it does not address the fundamental mechanism of artifact with SSFP cine MR, which is dependent on the shortest TR achievable by the gradients.
Cine MR with Non-Cartesian k-Space Sampling
As most commonly used, SSFP cine MR is implemented with Cartesian, or rectilinear, k-space sampling and cardiac gating. Other schemes, however, have also been described. Because radial k-space sampling seems likely to become more widely used in clinical practice, for angiographic as well as for cardiac applications, we will outline some of its more important properties below. In radial k-space sampling, radial projections replace the rectilinear phase-encoding steps (44). The following apply in the case of radial k-space SSFP cine MR:
1. In place of a phase-encoding gradient, the direction of the frequency-encoding gradient is rotated in a series of projections like the spokes of a wheel.
2. The sampling density is nonuniform; the outer portions of k-space are less densely sampled than the inner portions. All projections include the center of k-space—that is, the center is oversampled, and a filter must be applied to correct for this.
3. Whereas Cartesian undersampling produces an anatomically recognizable wraparound artifact, radial undersampling results in a linear streak artifact, which can have a bizarre appearance. This is because the undersampling occurs in the angular (azimuthal) direction rather than in the (constant) phase-encoding direction, as happens with Cartesian sampling (Fig 4a).
View larger version (14K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4a: (a) Diagram shows radial scanning. Data are acquired along radius of a circle. Multiple projections are acquired, with angular increment  k between projections. Distance between samples along radius kr defines field of view (FOV) in the frequency-encoding direction, 1/kr, as in standard Cartesian sampling. Note that FOV in the angular direction is given by 1/k, which is a function of radial distance from the center. So, angular FOV for high spatial frequencies is smaller than that for low spatial frequencies. However, because the center of k-space is sampled with every projection, full signal amplitude is acquired with all projections. This is different from Cartesian sampling and results in potentially high-signal-intensity streak artifacts, depending on the patient's geometry. kx = Cartesian k-space x-axis, ky = Cartesian k-space y-axis. (b) Short-axis SSFP cine MR images (2.2/1.1; flip angle, 45°) with radial k-space undersampling (top row) versus Cartesian k-space undersampling (bottom row). Characteristic wraparound artifact seen with Cartesian undersampling (arrows, bottom row) is replaced with streak artifact, characteristic of undersampled radial k-space acquisition (arrows, top row). (Adapted and reprinted, with permission, from reference 44.)
|
|
4. The pixel dimensions with radial sampling are determined only by the frequency-encoding gradient and are, in principle, isotropic. However, if there is angular undersampling (too few projections), the streak artifact will degrade image resolution.
5. Radial sampling is also associated with a moderate decrease in SNR, compared with the SNR in Cartesian sampling.
Real-time SSFP Cine MR Imaging
Although now most commonly used in its segmented k-space implementation, the original description of SSFP cine imaging was as a real-time ungated application (32). When the image matrix and spatial resolution are decreased, it is possible to shorten the TR further. In this mode, a low-resolution image can be acquired in less than 100 msec. When data acquisition is continuous, real-time imaging results, such that actual heartbeats are recorded, rather than "average" heartbeats, as is the case with a segmented acquisition. In this way, cine imaging can be performed without cardiac gating and, if needed, without breath holding (44,45) (Fig 4b). Although limited in spatial and temporal resolution in its current format, real-time cine imaging has potentially useful applications in cases of severe arrhythmia or when a gating signal is not available.
View larger version (166K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4b: (a) Diagram shows radial scanning. Data are acquired along radius of a circle. Multiple projections are acquired, with angular increment k between projections. Distance between samples along radius kr defines field of view (FOV) in the frequency-encoding direction, 1/kr, as in standard Cartesian sampling. Note that FOV in the angular direction is given by 1/k, which is a function of radial distance from the center. So, angular FOV for high spatial frequencies is smaller than that for low spatial frequencies. However, because the center of k-space is sampled with every projection, full signal amplitude is acquired with all projections. This is different from Cartesian sampling and results in potentially high-signal-intensity streak artifacts, depending on the patient's geometry. kx = Cartesian k-space x-axis, ky = Cartesian k-space y-axis. (b) Short-axis SSFP cine MR images (2.2/1.1; flip angle, 45°) with radial k-space undersampling (top row) versus Cartesian k-space undersampling (bottom row). Characteristic wraparound artifact seen with Cartesian undersampling (arrows, bottom row) is replaced with streak artifact, characteristic of undersampled radial k-space acquisition (arrows, top row). (Adapted and reprinted, with permission, from reference 44.)
|
|
As originally applied, real-time SSFP cine was performed with rectilinear k-space sampling. However, radial k-space variants have been described with very interesting and potentially useful properties. Larson and Simonetti (46) implemented echo-train imaging for accelerated real-time cine MR with an asymmetric, radial k-space sampling scheme. An extension of this work led to the implementation of "wireless cardiac gating" with radial k-space SSFP (47) for high-spatial-resolution cine imaging when a reliable ECG signal is unavailable. Recently, an alternative approach that utilizes spatial and temporal correlations in the cardiac MR signal has been described for real-time SSFP cine imaging; this technique also achieves improved temporal and spatial resolution (48). These latter techniques are not yet widely available, and their ultimate place in clinical practice, although very promising, remains to be defined.
SSFP Cine MR Imaging at 3.0 T
Recently, whole-body 3.0-T MR imaging systems have become available, and initial reports suggest they may have a role in cardiac imaging (49–51). There are, however, technical challenges for cardiac MR imaging at 3.0 T, particularly in relation to SSFP cine methods. Radiofrequency uniformity can be a problem at 3.0 T and, relative to 1.5-T imaging, radiofrequency power deposition increases by a factor of four. Therefore, specific-absorption-rate considerations limit the allowable flip angles and the minimum achievable TR. Also, the increased magnetic susceptibility effects at 3.0 T heighten sensitivity to off-resonance artifacts. These factors combine to make SSFP cine imaging more challenging at higher field strengths, because off-resonance (dark-band) artifacts are more troublesome. In the future, intravascular contrast agents may provide an alternative approach for cardiac cine imaging independent of SSFP techniques (52,53) (Fig 5). If the T1 of the blood is short, then spoiled GRE cine imaging with a very short TR becomes practical, with performance characteristics similar to those of SSFP. At the time of writing, however, no intravascular contrast agents have been approved by the Food and Drug Administration for cardiac imaging.
View larger version (104K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5a: Use of intravascular contrast agent at 3.0-T cine MR imaging in swine model. (a) Before contrast material injection, long-axis SSFP cine MR image (3.2/1.5 msec; flip angle, 55°; bandwidth, 900 Hz/pixel) shows considerable off-resonance artifact (arrows). (b) Also before injection, long-axis spoiled GRE cine image (3.1/1.7 msec; flip angle, 15°; bandwidth, 610 Hz/pixel) shows marked inhomogeneity as a result of blood saturation. (c) After injection of 0.1 mmol/kg Gadomer-17 (Schering, Berlin, Germany), image obtained with same sequence as in b shows uniformly high signal intensity for blood and excellent myocardial definition.
|
|
View larger version (146K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5b: Use of intravascular contrast agent at 3.0-T cine MR imaging in swine model. (a) Before contrast material injection, long-axis SSFP cine MR image (3.2/1.5 msec; flip angle, 55°; bandwidth, 900 Hz/pixel) shows considerable off-resonance artifact (arrows). (b) Also before injection, long-axis spoiled GRE cine image (3.1/1.7 msec; flip angle, 15°; bandwidth, 610 Hz/pixel) shows marked inhomogeneity as a result of blood saturation. (c) After injection of 0.1 mmol/kg Gadomer-17 (Schering, Berlin, Germany), image obtained with same sequence as in b shows uniformly high signal intensity for blood and excellent myocardial definition.
|
|
View larger version (142K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5c: Use of intravascular contrast agent at 3.0-T cine MR imaging in swine model. (a) Before contrast material injection, long-axis SSFP cine MR image (3.2/1.5 msec; flip angle, 55°; bandwidth, 900 Hz/pixel) shows considerable off-resonance artifact (arrows). (b) Also before injection, long-axis spoiled GRE cine image (3.1/1.7 msec; flip angle, 15°; bandwidth, 610 Hz/pixel) shows marked inhomogeneity as a result of blood saturation. (c) After injection of 0.1 mmol/kg Gadomer-17 (Schering, Berlin, Germany), image obtained with same sequence as in b shows uniformly high signal intensity for blood and excellent myocardial definition.
|
|
|
MYOCARDIAL TAGGING
|
|---|
Myocardial tagging refers to a family of techniques that lay out a saturation grid or series of saturation lines across the heart. Deformations of these lines due to myocardial contraction are then monitored (54). When combined with cine imaging, myocardial tagging can provide complementary or supplementary information about wall motion and is now usually performed as a breath-hold acquisition with a GRE readout (55).
Tagging can be used to measure myocardial strain in three dimensions, but quantitative analysis of tagged images is not straightforward. Recently, a fast simplified analysis of tagged images has been proposed (56), whereby strain is calculated by using the Fourier properties of the deformation pattern. In clinical practice, tagged images are usually analyzed subjectively to distinguish between normal and hypokinetic myocardial segments, and to evaluate regional circumferential contraction. A limitation of tagging is that the tag lines fade and the edges blur due to longitudinal relaxation (Fig 6). Tagging with SSFP imaging has recently been described, with advances in speed and SNR as compared with those of spoiled GRE tagging (57); however, clinical experience with this approach is limited at the time of writing. Although extremely powerful as a tool to study cardiac mechanics, the complexity of quantitative analysis limits the use of tagging to the subjective evaluation of wall motion, where it complements cine imaging very well. As a research tool, myocardial tagging has extensive applications.
View larger version (151K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6a: Short-axis breath-hold spoiled GRE myocardial tagging MR images (15/5; flip angle, 20°; acquisition time, 14 seconds; temporal resolution, 45 msec) obtained (a) 145, (b) 425, and (c) 700 msec after R wave. Note how distortion of tag pattern (arrow) gives insight into motion of myocardial wall. Note also that intensity of tag pattern fades over time, owing to longitudinal relaxation of tagged spins.
|
|
View larger version (152K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6b: Short-axis breath-hold spoiled GRE myocardial tagging MR images (15/5; flip angle, 20°; acquisition time, 14 seconds; temporal resolution, 45 msec) obtained (a) 145, (b) 425, and (c) 700 msec after R wave. Note how distortion of tag pattern (arrow) gives insight into motion of myocardial wall. Note also that intensity of tag pattern fades over time, owing to longitudinal relaxation of tagged spins.
|
|
View larger version (139K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6c: Short-axis breath-hold spoiled GRE myocardial tagging MR images (15/5; flip angle, 20°; acquisition time, 14 seconds; temporal resolution, 45 msec) obtained (a) 145, (b) 425, and (c) 700 msec after R wave. Note how distortion of tag pattern (arrow) gives insight into motion of myocardial wall. Note also that intensity of tag pattern fades over time, owing to longitudinal relaxation of tagged spins.
|
|
|
FLOW QUANTIFICATION
|
|---|
One of the attributes of the MR signal is that its phase can be used to encode flow information (58). Since its first introduction (21,59), gated flow quantification has been studied extensively in the heart and great vessels and has been applied successfully to valvular heart disease (60–63), congenital heart disease (64), and disease secondary to pulmonary hypertension (65–67). Although not limited by acoustic windows, flow quantification with MR imaging has features in common with pulsed Doppler ultrasonography in that it samples discretely and is therefore vulnerable to aliasing if the flow velocities are too high. This is because the phase of the MR signal can lie only in the range 0°–360°. So, a velocity that produces a phase shift of, for example, 370° owing to very fast flow is indistinguishable from one producing a phase shift of 10° owing to very slow flow. Velocity sensitivity with MR imaging can be adjusted so that aliasing does not occur (Fig 7). The velocity-encoding value (VENC) is defined as the velocity that produces a phase shift of 180°. If the highest velocity in the vessel of interest is less than the VENC, it can be measured without aliasing. If, however, flow velocity slightly exceeds the VENC, high-velocity flow in one direction may be erroneously encoded as fast flow in the opposite direction. Velocity encoding is accomplished by using gradients, and most commonly the direction of flow encoding is through plane—that is, in the section-selective direction. The optimal contrast-to-noise ratio in flow-encoding images is obtained when the flow velocity approaches the VENC. When the VENC is too high, although chances of aliasing are minimized, the contrast-to-noise ratio will be too low.
View larger version (155K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 7a: Normal aortic valve leaflet motility. Imaging was performed orthogonal to left ventricular outflow tract at the level of aortic valve cusps. (a) SSFP cine (3/1.5; flip angle, 60°; acquisition time, 8 seconds) and (b) phase velocity-encoding (56/2.6; flip angle, 30°; acquisition time, 15 seconds) MR images show normal valve area (arrow). (c) Graphs show flow velocity profiles corresponding to a (left) and b (right).
|
|
View larger version (128K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 7b: Normal aortic valve leaflet motility. Imaging was performed orthogonal to left ventricular outflow tract at the level of aortic valve cusps. (a) SSFP cine (3/1.5; flip angle, 60°; acquisition time, 8 seconds) and (b) phase velocity-encoding (56/2.6; flip angle, 30°; acquisition time, 15 seconds) MR images show normal valve area (arrow). (c) Graphs show flow velocity profiles corresponding to a (left) and b (right).
|
|
View larger version (13K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 7c: Normal aortic valve leaflet motility. Imaging was performed orthogonal to left ventricular outflow tract at the level of aortic valve cusps. (a) SSFP cine (3/1.5; flip angle, 60°; acquisition time, 8 seconds) and (b) phase velocity-encoding (56/2.6; flip angle, 30°; acquisition time, 15 seconds) MR images show normal valve area (arrow). (c) Graphs show flow velocity profiles corresponding to a (left) and b (right).
|
|
In the heart and great vessels, flow quantification may be performed with or without breath holding. Breath-hold acquisitions are fast and can prevent respiratory motion artifact, but the temporal resolution is generally limited to 50 msec or greater. Non–breath-hold flow quantification may achieve much higher temporal resolution but at the cost of increased imaging time and potential errors due to breathing motion artifact, unless respiratory gating is employed (68). It is important to exercise care both in the choice of VENC and in the orientation and offset of the imaging plane, which should be perpendicular to the vessel for through-plane flow VENC (Fig 8). Although the results of many studies testify to the reliability of flow quantification measures with MR imaging, the accuracy of the technique may be influenced substantially by eddy current effects, and regular calibration with a flow phantom is advisable.
View larger version (95K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 8a: Severe aortic stenosis. (a) Breath-hold SSFP cine MR image (3/1.5; flip angle, 60°) through aortic valve shows severe restriction (arrow) in leaflet motion (cf, Fig 6a). (b) Flow quantification image (56/2.6; flip angle, 30°) through the valve. (c, d) Pulsatile flow profiles show greatly increased peak flow velocity of 500 cm/sec (d). Maximum valve area, which is clearly shown and easily quantified with planimetry, was 0.6 cm2, reflecting critical stenosis.
|
|
View larger version (89K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 8b: Severe aortic stenosis. (a) Breath-hold SSFP cine MR image (3/1.5; flip angle, 60°) through aortic valve shows severe restriction (arrow) in leaflet motion (cf, Fig 6a). (b) Flow quantification image (56/2.6; flip angle, 30°) through the valve. (c, d) Pulsatile flow profiles show greatly increased peak flow velocity of 500 cm/sec (d). Maximum valve area, which is clearly shown and easily quantified with planimetry, was 0.6 cm2, reflecting critical stenosis.
|
|
View larger version (13K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 8c: Severe aortic stenosis. (a) Breath-hold SSFP cine MR image (3/1.5; flip angle, 60°) through aortic valve shows severe restriction (arrow) in leaflet motion (cf, Fig 6a). (b) Flow quantification image (56/2.6; flip angle, 30°) through the valve. (c, d) Pulsatile flow profiles show greatly increased peak flow velocity of 500 cm/sec (d). Maximum valve area, which is clearly shown and easily quantified with planimetry, was 0.6 cm2, reflecting critical stenosis.
|
|
|
MYOCARDIAL VIABILITY IMAGING: DELAYED CONTRAST ENHANCEMENT
|
|---|
The first clinical observations that myocardial infarcts enhance with extracellular contrast agents were made in the late 1980s (69–71). Early experiments in animal models showed clear enhancement of infarcts on T1-weighted spin-echo images, and the enhancement patterns in occlusive infarcts were noted to be different from those in reperfused infarcts (72). Enhancement of acute infarcts was noted to reach a peak at about 30 minutes after intravenous injection (70). Over the next several years, this phenomenon was studied in both animal and human models of acute and chronic infarction by several investigators (73–78), and the close topographic association between regions of infarction and regions of delayed hyperenhancement with gadolinium has been well established.
In recent years, myocardial viability imaging with delayed contrast enhancement has become widely accepted for the detection and characterization of acute and chronic myocardial infarction (76). The improved spatial resolution of MR imaging, compared with that of radionuclide imaging, provides clear advantages, particularly for nontransmural infarction, such that MR imaging is considered by some to outperform nuclear tomography (79–81). The precise mechanisms underlying contrast agent localization in nonviable myocardium are still somewhat controversial, but the phenomenological observation is straightforward. For an overview of the use of contrast agents in cardiac MR, the reader is referred to a recent publication by Edelman (82).
After intravenous injection, an extracellular tracer becomes distributed in the first pass according to myocardial blood flow and in the steady state according to extracellular fluid volume. Because myocardial scar is associated with a larger extracellular fluid volume than is viable myocardium, the steady-state concentration of gadolinium in scar is markedly higher than that in remote (noninfarcted) myocardium. In acute myocardial infarction, abnormal cell-membrane permeability allows free access of diffusible contrast material to the intracellular fluid space, and the apparent distribution volume of the agent may be even higher than that in the scar of chronic infarction. In either case, gadolinium-based agents tend to accumulate in nonviable myocardium within several minutes after intravenous injection and to wash out only slowly. The localization of such agents can be made conspicuous with an appropriate T1-weighted imaging technique, such as segmented k-space inversion-recovery GRE, as detailed by Simonetti et al (83) (Fig 9a). With magnitude reconstruction, the value of the chosen TI is important, because it determines the contrast between remote (viable) and infarcted myocardium. At the optimum TI, the signal from normal myocardium is nulled, while contrast-enhanced tissue is bright, having relaxed past the null point. The correct TI generally lies in the range of 200–300 msec, but this should be confirmed empirically on a patient-specific basis. If the TI is too short, normal myocardium will not be nulled and (with magnitude image reconstruction) positive signal will be generated because the myocardium has not yet relaxed to the null point. This may result in the so-called bounce artifact, where the signal from pixels with negative phase may be indistinguishable from those with positive phase, and pixels between adjacent positive and negative ones are dark (Fig 9b). To complicate matters further, the T1 (and therefore the optimum TI) of remote myocardium changes as the contrast agent is cleared from the blood pool.
View larger version (12K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 9a: (a) Diagram shows segmented k-space inversion-recovery turbo fast low-angle shot (FLASH) MR sequence. After R wave and suitable delay a spatially nonselective inversion pulse is applied, followed the TI. At this point, several lines (in this case, 23) of data are acquired during diastole. Next cardiac cycle is skipped. This pattern is repeated until required number of phase-encoding lines (a multiple of 23) is acquired. TI is generally chosen to null normal myocardium. (b) Effect of TI on regional myocardial signal intensities on 15 short-axis segmented turbo FLASH MR images (TR msec, echo time msec/TI msec, 8/4/110–450 [TI shown on each image]; flip angle, 20°) in a dog. Note transition from low to high signal intensity in anterior infarcted region (arrows) as TI increases. In this case, optimal contrast enhancement was achieved with 275-msec TI, when signal intensity of normal myocardium is nulled. (Reprinted, with permission, from reference 83.)
|
|
View larger version (112K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 9b: (a) Diagram shows segmented k-space inversion-recovery turbo fast low-angle shot (FLASH) MR sequence. After R wave and suitable delay a spatially nonselective inversion pulse is applied, followed the TI. At this point, several lines (in this case, 23) of data are acquired during diastole. Next cardiac cycle is skipped. This pattern is repeated until required number of phase-encoding lines (a multiple of 23) is acquired. TI is generally chosen to null normal myocardium. (b) Effect of TI on regional myocardial signal intensities on 15 short-axis segmented turbo FLASH MR images (TR msec, echo time msec/TI msec, 8/4/110–450 [TI shown on each image]; flip angle, 20°) in a dog. Note transition from low to high signal intensity in anterior infarcted region (arrows) as TI increases. In this case, optimal contrast enhancement was achieved with 275-msec TI, when signal intensity of normal myocardium is nulled. (Reprinted, with permission, from reference 83.)
|
|
To simplify the choice of TI, a phase-sensitive implementation of segmented inversion-recovery GRE MR imaging has been described (84). With phase-sensitive reconstruction, negative phase is assigned to the dark half of the gray scale and positive phase is assigned to the bright half. This eliminates the bounce artifact, which is sometimes a source of confusion on magnitude reconstructions and makes image contrast less dependent on accurate choice of TI. A drawback to phase-sensitive reconstruction is that background noise looks pixelated because its phase is random. In general, both phase-sensitive and magnitude images can be reconstructed from the same set of data, and if the TI is chosen appropriately evaluation of the magnitude image may be preferable. A method for determining the optimal TI is described below.
Both implementations of segmented inversion-recovery GRE MR are usually performed as breath-hold techniques, where one section is acquired in about 12 seconds and the whole heart is covered in about 10–12 sections (Fig 10). It is also possible to perform a single three-dimensional breath-hold acquisition to cover the entire heart, albeit with a longer breath hold. A free-breathing version of three-dimensional inversion-recovery GRE sequence has also been described (85), where navigator echoes from the diaphragm are used for respiratory gating. The various implementations of delayed contrast-enhanced inversion-recovery GRE work comparably well when used appropriately. The selection of the version to use may be a matter of personal choice, vendor platform, or availability.
View larger version (112K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 10a: Myocardial infarction. Breath-hold two-dimensional segmented inversion-recovery turbo fast low-angle shot MR images (2000/5/300; flip angle, 10°) obtained in vertical (a) long- and (b, c) short-axis orientations show subendocardial delayed hyperenhancement in anteroseptal (arrow) and inferoseptal (arrowhead) walls.
|
|
View larger version (107K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 10b: Myocardial infarction. Breath-hold two-dimensional segmented inversion-recovery turbo fast low-angle shot MR images (2000/5/300; flip angle, 10°) obtained in vertical (a) long- and (b, c) short-axis orientations show subendocardial delayed hyperenhancement in anteroseptal (arrow) and inferoseptal (arrowhead) walls.
|
|
TOP
ABSTRACT
INTRODUCTION
