DOI: 10.1148/rg.261055134
RadioGraphics 2006;26:275-297
© RSNA, 2006
AAPM/RSNA Physics Tutorial for Residents
MR Artifacts, Safety, and Quality Control1
Jiachen Zhuo, MS and
Rao P. Gullapalli, PhD
1 From the Department of Radiology, University of Maryland School of Medicine, 22 S Greene St, Baltimore, MD 21201. From the AAPM/RSNA Physics Tutorial for Residents at the 2004 RSNA Annual Meeting. Received June 27, 2005; revision requested September 16 and received October 13; accepted October 14. All authors have no financial relationships to disclose.
Address correspondence to R.P.G. (e-mail: rgullapalli{at}umm.edu).
 |
Abstract
|
|---|
Artifacts in magnetic resonance (MR) imaging result from the complex interaction of contemporary imager subsystems, including the main magnet, gradient coils, radiofrequency (RF) transmitter and receiver, and reconstruction algorithm used. An understanding of the sources of artifacts enables optimization of the MR imaging system performance. The increasing clinical use of very high magnetic field strengths, high-performance gradients, and multiple RF channels also mandates renewed attention to the biologic effects and physical safety of MR imaging. Radiologists should be aware of the potential physiologic effects of prolonged exposure to magnetic fields, acoustic noise, and RF energy during MR imaging and should use all the available methods for avoiding accidents and adverse effects. Imaging equipment should be regularly tested and monitored to ensure its stability and the uniformity of its functioning. Newly installed or upgraded MR systems should be tested by a physicist or qualified engineer before use. In addition, the authors recommend participation in the MR imaging accreditation program of the American College of Radiology to establish the initial framework for an adequate quality assurance program, which then can be further developed to fulfill local institutional needs.
© RSNA, 2006
|
Introduction
|
|---|
Since the invention of magnetic resonance (MR) imaging in 1972, the field has grown by leaps and bounds (1). Today, MR imaging is available in small community hospitals, in the far corners of the country, and in many private outpatient clinics. The large growth in this field is attributable to rapid technologic advances in several areas, including magnet technology, gradient coil design, radiofrequency (RF) technology, and computer engineering. In stride with the rapid technologic advances, there has been phenomenal growth in the number of applications for MR imaging. We no longer look to MR imaging to provide only structural information, but also functional information of various kinds. Information about blood flow, cardiac function, biochemical processes, tumor kinetics, and blood oxygen levels (for mapping of brain function) are just a few examples of the data that can be obtained with MR imaging today (2–6).
Since the first U.S. Food and Drug Administration (FDA)-approved MR imaging system arrived in the marketplace around 30 years ago, the available magnetic field strengths have increased 20-fold and more, the gradient capabilities have increased 100- to 200-fold, and the RF chain has been completely revamped with phased-array coil imaging and parallel imaging (7,8). We have gone from the single receiver to the quadrature coil, to multielement fast receivers. The growing demand for new applications has helped to fuel rapid growth in the number and type of pulse sequences available to radiologists. Imaging techniques have progressed from the simple spin-echo (SE) sequence to include steady-state, fast SE, echo-planar, and parallel imaging sequences (9,10). At the same time, radiologists and clinicians have had to adjust to the pace of change. They have had to learn to apply these new techniques and to distinguish between clinical features and potential artifacts on the resultant images.
Demands for higher spatial and temporal resolution for both structural and functional imaging have led to the proliferation of higher magnetic field strengths. Over the past 5 years, ultra-high-field-strength magnets (>1.5 T) have taken the market by storm and have quickly moved from a research environment to a clinical environment. The high magnetic field strengths and the high-performance gradients have brought a new awareness of the issue of safety for both clinicians and patients. The effects of exposure to magnetic fields and the compatibility of the many so-called MR-compatible or MR-safe surgical implants and other tools are being reinvestigated at high field strengths. The use of rapid imaging techniques such as echo-planar imaging and half-Fourier rapid acquisition with relaxation enhancement, or RARE, along with the use of ultra-high-speed gradients, has raised awareness and concern for the biologic effects of exposure.
With the expansion of the array of applications and increased number of users, it has become necessary to institute measures for quality control over the daily operations of MR imaging centers. The effort to control quality has been spearheaded mainly by the American College of Radiology (ACR), which has instituted a program for certification in MR imaging. Although certification is voluntary, more and more payers of clinical MR imaging costs are basing their payments on whether a site is ACR accredited or not.
The different components of an MR imaging system are shown in Figure 1. The main subsystems are the magnet, gradient coil, RF generator, and computer, the last of which controls the interplay between subsystems and the reconstruction, storage, and display of the images. There are many sources of artifacts on MR images. These could broadly be classified as image reconstruction–related, system-related, and physiology-related sources. Typically, image reconstruction–related artifacts occur because of limitations intrinsic to the reconstruction algorithm used by the particular vendor. System-related artifacts might be due to transient effects generated within one or more of the subsystems or could be a sign of degradation of some of the electronic components in the subsystem. System-related artifacts can be minimized through the adoption of a good quality assurance or quality control program, such as the one recommended by the ACR. Artifacts related to physiology are dependent on complex interactions between the subject and the MR imaging system. These include artifacts due to breathing-related or other motion of the subject, as well as flow-related artifacts. Physiology-related artifacts may be minimized with an understanding of the basic anatomy and physiology involved and the appropriate use of specific pulse sequences.
The next section, a brief overview of k-space, is followed by a discussion of the types of common artifacts seen on MR images. Safety concerns related to MR imaging are discussed according to subsystem, and guidelines for practice safety in the MR imaging workplace are provided. Finally, the authors suggest some basic quality assurance procedures that may help to minimize artifacts and increase system uptime while maintaining high quality.
|
k-Space
|
|---|
MR imaging systems collect data over time. The data that are captured during MR imaging are called k-space data or, simply, raw data. Typically, the data are collected by using quadrature detection, which provides both real and imaginary k-space data. k-Space data include useful information but can be interpreted only after they are translated into images with the Fourier transform method (Fig 2). Several types of motion may occur during the collection of k-space data. The motion may cause artifacts, which represent the complex interaction between the motion and the data acquisition time. Hence, an understanding of k-space is a critical element in the understanding and possible elimination of motion artifacts.
View larger version (60K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2. Schematic of quadrature coil detection shows the collection and combination of real and imaginary MR signal data to produce a complex map of k-space, which is then subjected to Fourier analysis to obtain the MR image.
|
|
The process of k-space detection and image acquisition is shown in Figure 2, and the typical characteristics of k-space are shown in Figure 3. As mentioned earlier, k-space is the rawest form of data obtained at MR imaging. An acquisition with a 256 x 256 matrix contains 256 lines of data, and each of those lines contains 256 data points. The y-dimension in this 256 x 256 array is called the phase encoding direction, and the x-dimension is called the frequency encoding direction. The distance between neighboring points in k-space determines the field of view of the object imaged, and the extent of k-space determines the resolution of the image. For a matrix with the same resolution, sparse sampling of the points would require less gradient strength and would produce images with a larger field of view, whereas faster sampling of points would require a higher gradient strength and would produce images with a smaller field of view. Filling of k-space is accomplished by acquiring frequency encoded data samples for a given phase encoding step. Sampling is much quicker in the frequency encoding direction than in the phase encoding direction, where the time between adjacent samples is greater than or equal to the repetition time of the particular sequence (except for fast SE and echo-planar imaging sequences). Low spatial frequencies are encoded in the center of k-space and provide contrast resolution to the image, whereas high spatial frequencies are encoded toward the edges of k-space and contribute to the spatial resolution of the image (11,12) (Fig 4). Fourier transform of k-space is then applied to convert the data into an image. Each pixel in the resultant image is the weighted sum of all the individual points in k-space. Therefore, the information in each pixel is derived from a fraction of every point in k-space. On the basis of these facts, we can conclude that any disruption of k-space, whether by motion, extraneous frequencies, or frequency spikes, has the potential to corrupt the entire image. In the next section, we describe the types of artifacts that might appear on MR images.
View larger version (92K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3. Schematic shows the location of low-frequency information about image contrast resolution at the center of k-space, and, at the edges of k-space, high-frequency information about spatial resolution and fine structure.
|
|
View larger version (31K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4a. Maps of k-space data (top) and corresponding MR images (bottom) obtained in a phantom illustrate the relationship between raw k-space data and the reconstructed MR image after Fourier transform. (a) Image reconstructed only with data from the center of k-space has high contrast but low spatial resolution (poor definition). (b) Image reconstructed only with data from the periphery of k-space shows well-defined edges but has poor contrast resolution. (c) Image reconstructed with all the k-space data has both good contrast and good spatial resolution.
|
|
View larger version (72K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4b. Maps of k-space data (top) and corresponding MR images (bottom) obtained in a phantom illustrate the relationship between raw k-space data and the reconstructed MR image after Fourier transform. (a) Image reconstructed only with data from the center of k-space has high contrast but low spatial resolution (poor definition). (b) Image reconstructed only with data from the periphery of k-space shows well-defined edges but has poor contrast resolution. (c) Image reconstructed with all the k-space data has both good contrast and good spatial resolution.
|
|
View larger version (70K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4c. Maps of k-space data (top) and corresponding MR images (bottom) obtained in a phantom illustrate the relationship between raw k-space data and the reconstructed MR image after Fourier transform. (a) Image reconstructed only with data from the center of k-space has high contrast but low spatial resolution (poor definition). (b) Image reconstructed only with data from the periphery of k-space shows well-defined edges but has poor contrast resolution. (c) Image reconstructed with all the k-space data has both good contrast and good spatial resolution.
|
|
|
Equipment-related Artifacts
|
|---|
Spike (Herringbone) Artifact
Gradients applied at a very high duty cycle (eg, those in echo-planar imaging) may produce bad data points, or a spike of noise, in k-space. The bad data might be a single point or a few points in k-space that have a very high or low intensity compared with the intensity of the rest of k-space. The convolution of this spike with all the other image information during the Fourier transform results in dark stripes overlaid on the image (Fig 5 ). The displacement of the spike of noise from the center of k-space determines the spacing between the stripes and the angulation of the stripes with respect to the readout direction. Spike noise usually is transient, but if it is not attended to, it can become chronic. Spike noise usually occurs because of loose electrical connections that produce arcs or because of the breakdown of interconnections in an RF coil, and it is more evident with the use of high-duty-cycle sequences (13).
View larger version (133K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5a. Spike artifact. Bad data points in k-space (arrow in b) result in band artifacts on the MR image in a. The location of the bad data points, and their distance from the center of k-space, determine the angulation of the bands and the distance between them. The intensity of the spike determines the severity of the artifact.
|
|
View larger version (166K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5b. Spike artifact. Bad data points in k-space (arrow in b) result in band artifacts on the MR image in a. The location of the bad data points, and their distance from the center of k-space, determine the angulation of the bands and the distance between them. The intensity of the spike determines the severity of the artifact.
|
|
On the other hand, the pattern of stripes produced by a spike in k-space can be used for tagging, an important technique in cardiac imaging. In this scheme, preparation pulses are applied prior to the imaging sequence. The preparation pulse has an excitation profile such that it forms echoes in different parts of the k-space. Fourier transform of such images produces tags in a grid-like pattern, with known spacing between the stripes. In cardiac imaging, these tags are applied at the start of each cardiac phase, and cardiac image acquisition then is performed at multiple phases of the cardiac cycle. Following the changes in tag position during the cardiac cycle from the point of initial placement can be useful for assessing cardiac motion (14).
Zipper Artifact
The zipper artifact is a common equipment-related artifact caused by the leakage of electromagnetic energy into the magnet room. It appears as a region of increased noise with a width of 1 or 2 pixels that extends in the frequency encoding direction, throughout the image series (Fig 6). All magnet rooms are shielded to eliminate interference from local RF broadcasting stations or from electronic equipment that emits an electromagnetic signal that could interfere with the MR signal. Leakage is usually caused by electronic equipment brought into the MR imaging room and the frequency generated by this equipment, which is picked up by the receive chain of the imager subsystem. The persistence of the problem even after the removal or electrical disconnection of all electronic equipment in the imager room might indicate that the RF shield has been compromised.
Motion-related Artifacts
Patient motion during image acquisition produces a commonly seen artifact. The artifact appears as blurring of the image as well as ghosting in the phase encoding direction. The time difference in the acquisition of adjacent points in the frequency encoding direction is relatively short (order of microseconds) and is dependent on the sampling frequency or the bandwidth used. The time difference in the acquisition of adjacent points in the phase encoding direction is much longer and is equal to the repetition time used for the sequence. The positional difference because of motion introduces a phase difference between the views in k-space that appears as a ghost on the image.
Respiratory and cardiac motion also can cause movement-related artifacts in the phase encoding direction. In general, periodic movements cause coherent ghosts, whereas nonperiodic movements cause a smearing of the image (15). Motion artifacts are generally caused by phase differences between adjacent k-space lines that are encoded at different phases of cardiac pulsation and respiration. Figure 7a shows an image affected by artifacts due to respiration. The best way to eliminate respiration-related motion is to perform breath-hold imaging (Fig 7b). However, the number of sections that can be obtained during a 20-second breath hold is limited. Multiple breath holds may be performed to increase coverage and obtain images in batches, but this requires cooperation from the patient. Some patients might have difficulty with breath holding. For such patients, respiratory gating may work better, with image acquisition performed at a certain phase of the respiratory cycle. The drawback of respiratory gating is that the acquisition time is increased. Another way to effectively reduce respiratory motion artifacts is to use respiratory compensation or phase reordering (also called respiratory-ordered phase encoding, or ROPE), whereby the phase encoding steps are ordered on the basis of the phase of the respiratory cycle (16). This technique allows a smooth phase variation across k-space and the limitation of variation to a single cycle of respiration instead of several cycles. Figure 8 shows an example of images acquired with and without respiratory motion compensation by using a fast SE sequence. Respiratory-ordered phase encoding works well in patients with regular respiratory cycles but not in those with irregular breathing.
View larger version (123K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 7a. Motion-related artifacts. (a) Abdominal image obtained without breath hold shows breathing-related motion artifacts. (b) Abdominal image obtained with breath hold shows a minimal artifact due to respiration and several motion artifacts due to cardiac pulsation.
|
|
View larger version (110K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 7b. Motion-related artifacts. (a) Abdominal image obtained without breath hold shows breathing-related motion artifacts. (b) Abdominal image obtained with breath hold shows a minimal artifact due to respiration and several motion artifacts due to cardiac pulsation.
|
|
View larger version (139K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 8a. Images acquired without and with compensation for respiratory motion. (a, c) Images acquired without compensation show respiration-induced artifact. (b, d) Images acquired with compensation are unaffected by respiratory motion.
|
|
View larger version (135K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 8b. Images acquired without and with compensation for respiratory motion. (a, c) Images acquired without compensation show respiration-induced artifact. (b, d) Images acquired with compensation are unaffected by respiratory motion.
|
|
View larger version (136K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 8c. Images acquired without and with compensation for respiratory motion. (a, c) Images acquired without compensation show respiration-induced artifact. (b, d) Images acquired with compensation are unaffected by respiratory motion.
|
|
View larger version (134K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 8d. Images acquired without and with compensation for respiratory motion. (a, c) Images acquired without compensation show respiration-induced artifact. (b, d) Images acquired with compensation are unaffected by respiratory motion.
|
|
Real-time navigator echo gating is an elegant non–breath-hold technique that can be used to compensate for several different types of motion (17) (Fig 9). With this technique, an echo from the diaphragm is obtained to determine the diaphragmatic position, and the timing of the acquisition is adjusted so that data are acquired only during a specific range of diaphragmatic motion. The navigator echo part of the acquisition is interleaved with the actual imaging sequence to facilitate real-time monitoring.
View larger version (168K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 9a. Placement of the navigator section for respiratory motion compensation. (a) Image with aqua overlay shows the navigator section from which the displacement information is obtained to determine the diaphragmatic position. (b) Graph shows diagphragmatic movement, indicated by the white wave and green line. The yellow boxes represent the best time to image ("window of opportunity").
|
|
View larger version (119K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 9b. Placement of the navigator section for respiratory motion compensation. (a) Image with aqua overlay shows the navigator section from which the displacement information is obtained to determine the diaphragmatic position. (b) Graph shows diagphragmatic movement, indicated by the white wave and green line. The yellow boxes represent the best time to image ("window of opportunity").
|
|
Cardiac pulsation can also lead to motion artifacts, especially during imaging of the heart. Cardiac motion–related artifacts are avoided by using electrocardiographic gating to time image acquisition so that it occurs at the same phase of each cardiac cycle. Further artifact suppression can be achieved while imaging the heart if the acquisition is performed during breath holding.
N/2 Ghost and Segmented-k-Space Artifacts
Over the past few years, single-shot and multishot imaging sequences have been used increasingly in the clinical environment. Echo-planar imaging is based on the continuous reversal of the echoes by using a gradient pulse (commonly called gradient-recalled echo [GRE] imaging) after a single excitation for the acquisition of all lines in k-space to form a single image. In this case, every alternate line in k-space is read in the opposite direction because of the reversed polarity of the readout gradient. Prior to Fourier transform of the k-space data, these lines must be reversed. This reversal process may result in the introduction of phase errors in every alternate line of k-space. Phase errors can arise from minor deviations in the linear course of the gradient as it traverses from maximum positive polarity to maximum negative polarity, the existence of eddy currents in the system, or poor shimming. The mismatch in phase causes the appearance of ghosts on the reconstructed images. On images acquired with a single-shot echo-planar imaging sequence, the ghost appears as an additional image with reduced intensity that is shifted by half the field of view, as shown in Figure 10b, since half the lines of k-space are different from the other half. This type of ghost artifact is commonly referred to as an N/2 ghost. The minimization of phase errors helps reduce the intensity of the ghost image while enhancing the main image (18). A similar phenomenon occurs in multishot imaging sequences in which there is a phase discrepancy between the segments of k-space. The number of ghosts on the image increases with the number of discontinuities in the k-space, as illustrated by Figure 10c. On images acquired with a multishot sequence, errors could result from the number of echoes per shot and the number of segments in the k-space. To accurately correct for or minimize these artifacts, it may be necessary to obtain an additional navigator echo (18,19).
View larger version (128K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 10a. Ghosts caused by phase errors at MR imaging in a phantom. (a) Initial image, acquired by using a fast SE sequence, shows no ghost artifact. (b) Image acquired with single-shot echo-planar imaging shows an N/2 ghost, a typical result of phase error (difference between odd- and even-numbered echoes). (c) Image acquired with a segmented-k-space MR sequence with eight phase-encoding lines per segment shows a coherent ghost.
|
|
View larger version (143K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 10b. Ghosts caused by phase errors at MR imaging in a phantom. (a) Initial image, acquired by using a fast SE sequence, shows no ghost artifact. (b) Image acquired with single-shot echo-planar imaging shows an N/2 ghost, a typical result of phase error (difference between odd- and even-numbered echoes). (c) Image acquired with a segmented-k-space MR sequence with eight phase-encoding lines per segment shows a coherent ghost.
|
|
View larger version (147K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 10c. Ghosts caused by phase errors at MR imaging in a phantom. (a) Initial image, acquired by using a fast SE sequence, shows no ghost artifact. (b) Image acquired with single-shot echo-planar imaging shows an N/2 ghost, a typical result of phase error (difference between odd- and even-numbered echoes). (c) Image acquired with a segmented-k-space MR sequence with eight phase-encoding lines per segment shows a coherent ghost.
|
|
Images acquired with fast SE sequences also are susceptible to segmented-k-space artifacts. Here the discontinuities may occur because of a minor discrepancy in the timing of the sequence (eg, between the multiple RF pulses or between the data collection windows) or because of eddy currents within the system. Any of these items could easily introduce phase errors that result in ghosting. In most situations, a user can do nothing but call service support to address the problem.
Flow Artifacts
Flowing blood is another source of motion-related artifacts, as shown in Figure 11. The blood flow is manifested as ghosting in the phase encoding direction. GRE sequences are much more susceptible to flow artifacts than are SE sequences. In SE sequences, the flow usually appears dark (produces no signal) because the flowing blood that is exposed to the 90° excitation pulse moves out of the imaging section before the refocusing 180° pulse is applied, while the blood that moved into the section at the same time was never exposed to the excitation pulse. In GRE imaging, the in-flow effect produces the bright-blood phenomenon. A common way to reduce the motion artifact caused by through-plane flow is to apply a saturation band adjacent to the imaging section. With this method, all spins within the slab are tilted toward the axial plane by a 90° RF pulse and then spoiled with the application of strong gradient crusher pulses before image acquisition starts. The spins, thus saturated, exhibit no signal when they move into the imaging volume (Fig 11b ).
View larger version (119K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 11a. Control of flow-related artifacts. (a) Image shows a cardiac pulsation artifact. (b, c) Images obtained with a saturation band superior to the imaging section (b) and with first-order motion compensation (c) show less severe flow-related artifacts.
|
|
View larger version (116K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 11b. Control of flow-related artifacts. (a) Image shows a cardiac pulsation artifact. (b, c) Images obtained with a saturation band superior to the imaging section (b) and with first-order motion compensation (c) show less severe flow-related artifacts.
|
|
View larger version (120K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 11c. Control of flow-related artifacts. (a) Image shows a cardiac pulsation artifact. (b, c) Images obtained with a saturation band superior to the imaging section (b) and with first-order motion compensation (c) show less severe flow-related artifacts.
|
|
Flow artifacts could also be minimized by using flow compensation or gradient moment nulling. In this technique, flowing spins are rephased by using motion-compensating gradient pulses. As illustrated in Figure 12, if no motion compensation is applied with the gradients, the flowing spins are not in phase with the static spins when the echo forms. With motion compensation, the flowing spins with constant velocity (irrespective of what the velocity is) are brought back into phase, with no effect on static spins. Figure 11c illustrates the decrease in the severity of flow artifacts with the use of first-order motion compensation, in comparison with flow artifacts in Figure 11a, which was acquired without motion compensation. The penalty for using motion compensation is increased echo time. Higher-order motion compensation also can be obtained for spins with constant acceleration and for spins with varying acceleration (wobbling) by applying additional gradient pulses, but this increases the echo time even further (20).
View larger version (18K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 12a. Schematics show the MR signal effects of flow compensation with gradient moment nulling. Top: Typical readout gradient waveform. Bottom: Phase of cumulative stationary spins (solid line) and constant-velocity spins (dotted and dashed lines). (a) During imaging without flow compensation, the moving spins are not refocused at the desired echo time (TE), and this leads to the loss of signal from flowing spins. (b) During imaging with gradient moment nulling, all the spins are refocused, and flow velocity is compensated for by the 1:2:1 ratio of the gradient lobe areas. GGMR = gradient moment nulling gradient, GR = readout gradient.
|
|
View larger version (26K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 12b. Schematics show the MR signal effects of flow compensation with gradient moment nulling. Top: Typical readout gradient waveform. Bottom: Phase of cumulative stationary spins (solid line) and constant-velocity spins (dotted and dashed lines). (a) During imaging without flow compensation, the moving spins are not refocused at the desired echo time (TE), and this leads to the loss of signal from flowing spins. (b) During imaging with gradient moment nulling, all the spins are refocused, and flow velocity is compensated for by the 1:2:1 ratio of the gradient lobe areas. GGMR = gradient moment nulling gradient, GR = readout gradient.
|
|
Susceptibility Effects
When placed in a large magnetic field, tissues are temporarily magnetized, with the extent of the magnetization depending on the magnetic susceptibility of the tissue. The effect of tissue magnetization slightly alters the local magnetic field. The difference in tissue susceptibility causes field inhomogeneity between tissue boundaries, which makes spins dephase faster, producing signals of low intensity. This signal loss is especially severe at air-tissue or bone–soft tissue boundaries, because air and bone have much lower magnetic susceptibility than do most tissues. Local magnetic fields tend to take on different configurations around these interfaces, and the difference introduces geometric distortion into the resultant images, especially when sequences with long echo times are used. SE sequences are less affected by local field inhomogeneity because of the 180° refocusing pulse, which cancels the susceptibility gradients. Echo-planar images generally are subject to more severe susceptibility artifacts because the echoes are refocused by using gradient refocusing over a long period. Figure 13a shows a susceptibility artifact that occurred at echo-planar imaging. Note the distortion, near the air-tissue interface, that was caused by the susceptibility difference. One way to minimize this distortion is to orient the phase encoding gradient along the same axis as the susceptibility gradients. Figure 13a shows echo-planar images acquired with the phase encoding gradient in a left-right direction with regard to the image, whereas Figure 13b shows the same echo-planar images acquired with the phase encoding gradient in the anterior-posterior direction. Although the severity of the distortion is markedly reduced, there is still some residual distortion and some stretching in the direction of the phase encoding gradient. A good understanding of the anatomy and the types of tissues involved may help reduce susceptibility-related artifacts. The best way to minimize susceptibility artifacts is to use an SE sequence or reduce the echo time and increase the acquisition matrix. Proper shimming over the volume of interest before image acquisition is also critical to improve the local field homogeneity.
View larger version (157K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 13a. Echo-planar images show magnetic susceptibility artifacts. (a) Left-right phase encoding causes severe distortion of the signal (arrows). (b) Anterior-posterior phase encoding minimizes the distortion, since the phase axis is symmetric around the susceptibility gradients.
|
|
View larger version (158K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 13b. Echo-planar images show magnetic susceptibility artifacts. (a) Left-right phase encoding causes severe distortion of the signal (arrows). (b) Anterior-posterior phase encoding minimizes the distortion, since the phase axis is symmetric around the susceptibility gradients.
|
|
The artifact that is caused by the presence of metallic implants is of the same nature, but it is more severe because most metals have much higher magnetic susceptibility than does body tissue. As shown in Figure 14, metal-related artifacts typically are manifested as areas of complete signal loss because the local magnetic field is so strong that the spins are almost immediately dephased. Sometimes, depending on the strength of the resultant distortion of the magnetic field, part of a section may appear on images acquired in a completely different section because of the frequency difference between the spins in the metal and those next to the metal. The visual distortion created by metallic implants on MR images cannot be completely eliminated, but its effect can be minimized by a large receiver bandwidth and a decreased echo time. Fast SE acquisitions with a high bandwidth typically work well in such cases. However, care should be taken to avoid the heating of tissue that is adjacent to the metal (21).
Chemical Shift Artifacts
Protons in water and protons in fat have a significantly different chemical environment, which causes their resonance frequencies to differ. At 1.5 T, protons from fat resonate at a point approximately 220 Hz downfield from the water proton resonance frequency, and this difference in frequency is linearly related to magnetic field strength. The shift in Larmor frequency between water protons and fat protons is referred to as chemical shift. The suppression of fat protons at MR imaging is very useful for improving contrast on images of the breast, abdomen, and optic nerve (22).
The chemical shift between water and fat can cause artifacts in the frequency encoding direction. If this occurs, there will be a slight misregistration of the fat content on images, because of the slight shift in the frequency of the fat protons. The number of pixels involved in this slight shift depends on the receiver bandwidth and the number of data points used to encode the frequency direction. In mathematic terms, the result may be expressed as follows: CSA = 
• Nfreq/BWrec, where CSA is a chemical shift artifact, is the frequency difference between fat and water, Nfreq is the number of samples in the frequency encoding direction, and BWrec is the receiver band-width. This implies that the chemical shift can be minimized by using a higher receiver bandwidth.
Chemical shift artifacts were common on T2-weighted images obtained before the invention of fast SE imaging. The overall signal intensity on a normal T2-weighted image was low, and in order to increase the signal-to-noise ratio, it was a common practice to use a low receiver bandwidth for the T2-weighted echo. With the use of fast SE sequences, in contrast, the goals are to minimize the echo spacing and to maximize the decaying T2 signal. These goals are usually achieved by using a shorter RF pulse and a higher-bandwidth receiver window.
The signal from fat can dominate the dynamic range of an image and suppress the contrast between lesions and normal tissue, especially on abdominal MR images. Echo-planar images are very susceptible to artifacts from off-resonance frequencies, and unless the fat frequency is suppressed, the images will be degraded by fat-related artifacts. Any off-resonance signal (such as that of fat) will be affected by a position shift along the image because of the long duration of data sampling (50–100 msec). Such a shift could affect approximately 12 pixels at a readout time of 50 msec, as shown in Figure 15a. A common method for minimizing fat-related artifacts is to apply a frequency-selective RF pulse to null the fat signal before applying the imaging pulse sequence (23). Successful fat saturation will result only if the magnetic field homogeneity is sufficient throughout the imaging region. Figure 16 illustrates the difference between a poorly shimmed object and a well-shimmed object at MR imaging with fat saturation. The result, in the case of the poorly shimmed object, is saturation not only of fat but also of a significant amount of water content; the resultant images are nondiagnostic. Therefore, it is necessary to properly shim the region before applying fat saturation. Proper shimming results in the depiction of two distinct signals in fat and in water. As shown in Figure 15b, good shimming, followed by fat saturation, enables the avoidance of off-resonance effects from fat. Another method is to use a short inversion time inversion recovery (STIR) sequence (24), as shown in Figure 17. In this method, the short T1 of the fat signal is used to suppress the signal from fat through the use of an inversion recovery sequence. If the T1 of any tissue is known, the inversion time needed to null the signal from that tissue in an inversion recovery sequence is given by 0.7 • T1. In the case of fat, the inversion time is around 150 msec at 1.5 T. The decision to use fat saturation or a STIR sequence depends on the specific application; both methods have advantages and disadvantages (25).
View larger version (117K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 15a. Chemical shift artifact at echo-planar imaging. (a) Image shows severe chemical shift artifact from insufficient fat suppression. (b) Image obtained with fat saturation shows minimization of the chemical shift artifact or off-resonance effect.
|
|
View larger version (104K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 15b. Chemical shift artifact at echo-planar imaging. (a) Image shows severe chemical shift artifact from insufficient fat suppression. (b) Image obtained with fat saturation shows minimization of the chemical shift artifact or off-resonance effect.
|
|
View larger version (17K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 16. Schematics show the effect of fat saturation with an RF pulse applied at different frequencies for differentiation of water from fat. Left: Good magnetic field homogeneity, which results from a good shim, ensures effective fat suppression on images. Right: Poor homogeneity, or a bad shim, may produce nonuniform fat suppression on images and compromise the clinical reading.
|
|
View larger version (17K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 17. STIR sequence. The inversion time (TI) is set so that the available net magnetization of fat is zero at 0.7 • T1 of fat. ADC = analog-to-digital converter, Mz = longitudinal magnetization, RF = RF pulse.
|
|
In-phase versus Opposed-phase Imaging
Chemical shift can be used to advantage, as is routinely done in liver imaging for the detection of fatty infiltration. Because the precessional frequencies of fat and water are different, they go in and out of phase from each other after excitation. Since the difference in frequency between fat and water is about 220 Hz, the fat and water signals are in phase every 1/220 Hz. In other words, they are in phase at 4.45 msec, 8.9 msec, and so on (these time points correspond to the first cycle, second cycle, and later cycles). Within each cycle, the fat and water signals would be 180° out of phase with one another at 2.23 msec, 6.69 msec, and so on. Figure 18 shows the difference between in-phase and opposed-phase abdominal images obtained with echo times of 2.2 msec and 4.4 msec, respectively, at 1.5 T. At the in-phase echo times, the water and fat signals are summed within any pixel that contains a mixture of the two components, whereas at opposed-phase echo times, the signal in these pixels is canceled out, leading to the appearance of a dark band at the fat-water interface. In-phase and opposed-phase imaging sequences are sometimes also referred to as chemical shift sequences (26).
Aliasing (Wraparound) Artifacts
A common artifact routinely encountered in MR imaging is aliasing. Aliasing occurs whenever the imaging field of view is smaller than the anatomy being imaged. It should be noted that the field of view in both the readout and the phase encoding directions is inversely proportional to the incremental gradient step required from one point to the next. This means that the field of view is the distance along the gradient that one must move to experience one complete cycle. Therefore, when a field of view is selected, it is expected that the gradient will move one cycle (0–2
) from one end of the field to the other. If the sensitivity of the RF transmission coil extends beyond this field of view, the spins outside the field will be excited, but they will be part of the next cycle in the phase encoding direction (eg, 2–4, 4–6, and so on). However, the phase angles of the spins outside the field of view are essentially equivalent to those of the spins within the selected field of view but on the opposite side of the image. For the Fourier transform, spins at 10° are equivalent to spins at 2 + 10°, and the result is an overlap of signals outside the intended field of view with signals within that field of view. Figure 19 shows aliasing of signals from outside the field of view in the imaging volume of interest. Aliasing is always encountered in the phase encoding direction because the frequency encoding direction is typically oversampled within the system. Aliasing also can occur in the section direction in a three-dimensional (3D) acquisition. In this case, the side lobes of the RF pulse that excites the 3D slab produce a signal from outside the field of view, and information in the first section may be aliased in the last section of a 3D acquisition.
View larger version (145K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 19a. Aliasing artifact caused by imaging with too small a field of view. (a) Image shows aliasing artifact. (b) Image obtained with the phase encoding and frequency encoding axes exchanged shows no aliasing.
|
|
View larger version (140K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 19b. Aliasing artifact caused by imaging with too small a field of view. (a) Image shows aliasing artifact. (b) Image obtained with the phase encoding and frequency encoding axes exchanged shows no aliasing.
|
|
Especially in the coronal plane, large fields of view that are smaller than the object will not only cause aliasing but also create an interference pattern called the moiré or fringe artifact. Homogeneity of the main field over large fields of view degrades toward the edges of the field, causing phase differences between the two edges. When aliasing occurs, the overlap of signals from one side of the body to the other side, with mismatched phases, produces a moiré artifact (Fig 20).
An easy way to eliminate aliasing is to exchange the readout direction with the phase encoding direction (swap the imaging axes) so that the anatomy fits into the phase field of view. However, this method may result in motion artifacts along the phase encoding direction that obscure a pathologic entity. Other ways of decreasing aliasing artifacts are to increase the phase field of view or to apply spatial saturation pulses outside the field of view so that the signals from outside the field of view will not give rise to artifacts within the field.
The interference pattern also occurs when different echoes from different excitation modes are received within the same acquisition window but with slightly different echo times. The second echo is often a stimulated echo. The spacing of the fringes in this case is inversely proportional to the difference in echo times. The only way to address this problem is either to adjust the timing of the sequence so that the stimulated echo is superimposed on the main echo or to use selective saturation to null the signal from the stimulated echo.
Gibbs Effect
Gibbs or truncation artifacts are bright or dark lines that appear parallel with and adjacent to the borders of an area of abrupt signal intensity change on MR images. Insufficient collection of samples in either the phase encoding direction or the readout direction leads to Gibbs rings as a result of the Fourier transform (27) (Fig 21). One usually sees Gibbs rings in the phase encoding direction because a phase encoding matrix that is smaller than the readout matrix is often selected to decrease the acquisition time. Gibbs rings can be minimized by increasing the acquisition matrix size and maintaining the same field of view, but the improvement comes with a penalty of increased acquisition time and a reduced per-pixel signal-to-noise ratio.
View larger version (24K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 21a. Gibbs ring artifact. (a) Axial image obtained with a low spatial resolution (128 x 128) in a cylinder shows a Gibbs ring artifact at the edges of the cylinder. (b) Image obtained with a higher spatial resolution (256 x 256) shows minimization of the artifact. The dotted line indicates the desired object profile, and the red line indicates the object profile with two different resolution parameters.
|
|
View larger version (22K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 21b. Gibbs ring artifact. (a) Axial image obtained with a low spatial resolution (128 x 128) in a cylinder shows a Gibbs ring artifact at the edges of the cylinder. (b) Image obtained with a higher spatial resolution (256 x 256) shows minimization of the artifact. The dotted line indicates the desired object profile, and the red line indicates the object profile with two different resolution parameters.
|
|
Section Cross Talk
Cross talk between sections occurs at imaging of contiguous sections (with no gap) in a multisection two-dimensional acquisition. Typically, in the presence of cross talk, all sections except the edge sections in a multisection acquisition have reduced signal intensity. The RF pulses used for excitation are not perfect in that they do not produce a section profile with a straight edge. In addition, any side lobes to the RF pulse might excite the neighboring section by a few degrees. In other words, the neighboring section is subjected to an RF pulse more than once during a single repetition time, which causes partial saturation of the signal in that section and leads to a lower signal intensity. Possible remedies for cross talk between sections include RF pulses with sharper section profiles, an increased gap between sections, and/or multiple sections imaged in separate batches (one batch consisting of all the odd-numbered sections; the other, all the even-numbered sections). Three-dimensional images are not vulnerable to this effect because the whole volume undergoes excitation and because sections within the volume are acquired by using the gradients, as is the case with normal frequency and phase encoding.
Imaging System Cross Talk
Many imaging centers have more than one MR imaging system with the same field strength. If multiple MR imaging systems are operated at about the same frequency and at the same time, it is very important that each of the MR imaging rooms be surrounded by a tight shield to avoid cross talk between the imagers. It is probably best to ground the RF shield separately and keep the RF and gradient amplifiers for the individual imagers away from each other. To ensure that cross talk is not occurring, regular testing should be performed by applying intensive RF pulses and high-bandwidth reception on both imagers simultaneously.
Gradient-related Distortion
When an electrical current is applied to a gradient coil, a varying magnetic field is produced that is linear through the isocenter of the magnet but tapers toward the sides of the magnet. As illustrated in Figure 22, the gradient linearity gets worse with the distance from the isocenter. The effect of this distortion is to compress the images, thereby mismapping the spins from their true location at the edges of images obtained with a large field of view. Most MR imagers are equipped with distortion-correction algorithms to compensate for the gradient nonlinearity. Figure 23a shows the distortion on a spinal image with a large field of view, and Figure 23b shows the effect of the distortion-correction algorithm on the image. Because magnet sizes have become smaller in recent years, the possibility of geometric distortion should be given greater consideration, and the accuracy of correction algorithms should be assessed (28).
View larger version (12K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 22. Schematic shows the geometric distortion of a typical gradient profile along the x-axis, with decreasing linearity (solid line) as the distance from the magnet isocenter increases. The red dotted line shows the desired linear gradient profile. Bz = magnetic field strength, Max FOV = maximum field of view.
|
|
View larger version (96K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 23a. (a) SE image obtained with a large field of view shows the result of gradient geometric distortion. (b) Image obtained with a vendor-supplied correction algorithm shows correction of the geometric distortion.
|
|
View larger version (98K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 23b. (a) SE image obtained with a large field of view shows the result of gradient geometric distortion. (b) Image obtained with a vendor-supplied correction algorithm shows correction of the geometric distortion.
|
|
Parallel Imaging Artifact
New signal acquisition schemes such as SMASH (simultaneous acquisition of spatial harmonics), SENSE (sensitivity encoding), and GRAPPA (generalized autocalibrating partially parallel acquisitions), which are categorized as parallel imaging techniques (29–31),
Top
Abstract
Introduction
