 |
 |
 |
 |
 |
 |
 |
Carl's Roost:MRI Artifact Gallery 
This is a collection of MRI images demonstrating some common and not so-common artifacts. Many of the artifacts occur to some extent in more than one image, there are multiple names for some types of artifacts, and some images illustrate multiple problems. The following list is provided as an index: Raw data (interferograms) are also displayable in some cases. These are typically just the real part of the complex raw data, and have been scaled to emphasize the interesting details. Therefore, they cannot be successfully transformed to reproduce the displayed images. In some cases, the original data file may be available for downloading and further investigation. If this file is available, it is in the Surrey Medical Imaging Systems (SMIS) ".MRD" format. All such files should be downloaded or saved to disk as "raw", "binary" or "source", depending on your browser. Note: The appearance of the images will depend strongly upon your browser, and monitor or printer settings. Some of the artifacts are subtle enough that a poor display setup may mask them completely. You can download the image and view it off line with better software, if necessary. Acknowledgements: This collection was started during a Biophysics 404 class at the University of Illinois in Fall 1994. Thanks to Jill Hanson for compiling the original list. This work is supported by the U.S. National Institutes of Health, Center for Research Resources, and the University of Illinois. Experimental details are at the end. Reference Image
This image is as close to "ideal" as we could get. Compare each of the other images to this. Note that the position of the phantom does change in some of the images. Interferogram and raw data file available. Back to Top Arcing - Loose Connections
These images demonstrate intermittent corruption of the data, typically by electrical problems. They have been artificially generated by using viewit to modify the raw data from the reference image before transforming it. On the left, a few data points have been modified, as might happen if an electrical transient (such as a static discharge, sometimes called an "arc") occured during the digitization of one echo. Note the elevated background level. The venetian blind pattern becomes a crosshatch with two arcs, and with multiple arcs, may be indistinguishable from random noise. Note that a transmitter arc looks different, since it affects the MR magnetization directly, and does not occur during digitization. See the instability example for what might appear in that case. Interferogram here. You'll have to look very closely to see the spike, in the upper left quadrant. On the right, some part of each of three rows of the data has been set to zero, as might happen if a computer problem caused loss of data, or a loose cable momentarily (for a few milliseconds) prevented the received signal from being digitized (sometimes called "dropout"). Note the similarity (and differences) to the motion/instability case. The width of the band at the center is inversely proportional to the duration of the dropouts. Interferogram here. Look closely to see the droputs. Back to Reference Image or Top Audio Frequency Problems
These two images demonstrate two types of audio-frequency problems (in this case, at AC power-line - "mains" - frequency, 60 Hz and harmonics, especially 300 Hz):
Problem #1 shows up as ghosts, weak copies of the real image, displaced along the phase-encoding axis. The number and intensity of the ghosts depends upon the relationship between the period of the audio modulation and the repetition time Tr. Note that this is similar, but not identical, to the general instability case shown below, where uncorrelated effects cause a general smearing, rather than discrete ghosts. Problem #2 shows up as lines or spots at the appropriate points along the frequency axis. Lines are generated if there is no correlation between the audio period and Tr, otherwise, discrete spots occur. In this case, the frequencies are +/- 300 Hz. Since the spectral width is 50 KHz and there are 256 points along the frequency axis, the spots or lines are about 3 pixels apart. Notice that changing the rf reference frequency by 3900 Hz causes the image to move (since it is a real radio-frequency signal) but has no effect on the position of the artifact (since it is not present at the radio-frequency). Both problems can be mitigated by use of AC-line synchronization (line trigger). This results in all of the intensity in these two columns being collapsed to the central line of the image. See also the discussion of the DC artifact, and the chemical shift example. Back to Reference Image or Top Chemical Shift
A cross-section of a raw egg demonstrates the chemical shift artifact. Both images were acquired with Te/Tr 400/30, but two different readout gradients. On the right, a 50kHz bandwidth and 16 averages, on the left, 12.5kHz bandwidth and 1 average (averaging adjusted to keep signal-to-noise levels constant). Note the apparent displacement of the yolk image in the left image. This is because the yolk is principally fat (see the egg chemistry discussion). The fat has a chemical shift ranging from about +100 to -700 Hz relative to the water of the egg white. 700 Hz for a 12.5 kHz bandwidth over 256 pixels corresponds to a shift of 14 pixels. The diffuse border of the fat area is due to the variable chemical shift of the lipids. The narrow bandwidth image also displays a particularly severe case of audio frequency artifacts. These are spread out over several pixels due to the narrow bandwidth, and the presence of multiple harmonics of 60 Hz (principally 180 and 300 Hz). The FID artifact is also present in this image, as a line at the top, showing the use of phase cycling to avoid corruption of the image itself. The FID disappears in the wide-bandwidth image due to signal averaging. Back to Reference Image or Top Clipping - ADC overflow
The receiver gain was increased (by 20 dB) so that the signal level was larger than the maximum ADC (Analog-to-Digital-Converter, or digitizer) value. To compensate, the reconstructed image brightness was scaled by x0.1 relative to the others. This effect is called "clipping" because on a plot of signal amplitude vs. time, it looks like the top and bottom of the echo has been "clipped off" with scissors. Note the overall intensity loss as well as the extensive signal reconstructed outside the object. Raw data here. Back to Reference Image or Top DC Offset
Actually, if you look carefully, all of the images in this collection have this artifact - a single pixel at the exact center which is brighter or darker than the surroundings. This is the Fourier transform of a constant offset in the raw data. This is extremely common, since temperature fluctuations often cause dc-coupled amplifiers to have non-zero outputs with zero input. Normally it is compensated for in software, by applying a "dc correction" or "baseline correction" before the FT. To produce the example, the reference data was reprocessed using only 128 phase encodings. Because zero-filling is employed before the FT, a Gibbs artifact results in the spot "ringing" into adjacent pixels. See also the Gibbs example. Note: do not confuse the two pixels on either side of center with this DC artifact. These two bright spots also appear in most of the collection, and result from an audio (line) frequency artifact (see above) which is perfectly synchronized with the acquisition, and hence reconstructs to the center of image space. The three pixels together make a tiny "hyphen". Note that the zero-filled (Gibbs) example does not show these adjacent pixels ringing: this is because their intensity is much closer to the average image intensity than the that of the DC spike. This artifact should not appear in a true "digital" receiver which has IF digitizing and performs quadrature detection numerically. Back to Reference Image or Top Digitizer Quantization - Low Dynamic Range
The original data set was processed (with viewit) to reduce the precision. Originally digitized at 16 bit resolution (values +32767 to -32768), the lower 8 bits were all set to zero, resulting in an effective digitizer precision of only 8 bits (+127 to -128). The noise level is smaller than the digitizer quantum in this case, so we see the effects of quantization. No raw data available. Back to Reference Image or Top FID (with spin echo)
A combination of B1 inhomogeneity, poor slice profile, and insufficient spoiler gradients between the refocussing pulse and the readout interval of a spin-echo sequence results in an FID signal being detected along with the echo. Since the FID is not phase encoded (normally the phase encoding occurs before the refocussing pulse), it is not dispersed along the phase encoding axis, but appears as a line across the center of the image. This example is a very mild case. Receiver turn-on artifacts (not true NMR signals) appear similarly, except that they extend into the signal-free region, while the FID (a real NMR signal) is confined to the projection of the sample along the readout axis. Proper phase cycling can relegate this artifact to the edge of the image. This is demonstrated in the chemical shift example. The FID can be seen at the left edge of the interferogram. The reference interferogram was acquired with phase alternation. See if you can see the difference. FID artifacts cannot occur on gradient echo images, but receiver turn-on is still possible on gradient echoes. See shimming for an example of this. Back to Reference Image or Top Gibbs Ringing - Truncation
Only 64 samples were acquired in the phase encode direction. Also, although 256 samples were acquired in the readout direction, 192 of them were discarded, keeping only the central 64 for reconstruction. In addition to the expected blurring (loss of resolution) there are low intensity "rings" parallel to the edges of the sample. Contrast enhanced slightly relative to other images in this series. Raw data here.
Back to Reference Image or Top Moire
When echoes from different excitation modes occur within the acquisition window, but with slightly different echo times, an interference pattern occurs. This looks like the pattern of superimposed screens, known as a Moire' pattern, and can be seen at the bottom of the image above. The second echo is often a stimulated echo, or a multiple-quantum echo. The spacing of the fringes is inversely proportional to the difference in echo timings. This artifact is often sensitive to shimming or suxceptibility gradients. This image was acquired with a shorter Tr (200msec, less than T2 for this sample) and with the "spoiler" or "crusher" gradient at the end of the sequnece turned off. Receiver gain was increased to compensate for T1 losses. Contrast considerably enhanced, emphasizing some B1 inhomogeneity. Sometimes both echoes can be seen in the interferogram, in this case the second echo is in the upper right quadrant. Raw data here. Back to Reference Image or Top Motion - Instability
The sample in the left image was jostled with a stick during the acquisition. Virtually any instability or fluctuations in the system will cause artifacts of this type, of greater or lesser intensity. This includes power supply problems, mechanical vibrations, cryogen boiling, large iron objects moving in the fringe field, loose connections anywhere, or pulse timing variations, as well as sample motion. The displacement need not be very much, as can be seen from the absence of any shift along the frequency axis. Raw data here. A much more extreme example is shown on the right, where one of the shim currents was changed drastically several times during the acquisition. This results in a frequency shift, as well. This particular case is much less common, but motion along the frequency axis can look similar. Back to Reference Image or Top Quadrature Ghost
The gain of the imaginary receiver channel was changed slightly, prior to the aquisition. (Phase errors between the two quadrature channels can also cause the same effect.) Note that the ghost is displaced diagonally across the center in both readout and phase encoding directions, in contrast to almost all other types of ghosts. Contrast enhanced. Raw data here. This artifact should not appear in a true "digital" receiver which has IF digitizing and performs quadrature detection numerically. Back to Reference Image or Top Radio Frequency Interference (RFI).
The interfering signal is at 170.000000 MHz, while the center frequency of the images is at 170.029470 MHz. The left image has a 50 kHz spectral width. The interference is thus 4470 Hz beyond the left (low-frequency) edge of the image, and is attenuated and aliased to the high frequency side. The right image has a 100 kHz spectral width. This artifact is thus within the receiver bandwidth, and is much stronger. In the second image, the interference is strong enough to be visible as a horizontal herringbone pattern in the interferogram. Raw data here. Back to Reference Image or Top Random Noise
There are various noise sources in any electronic system, including Johnson noise, shot noise, thermal noise, etc. These are more or less indistinguishable in images. For this image, the slice thickness was reduced by a factor of 10 (to 1mm) to reduce the signal 10-fold. The receiver gain was increased by a factor of 10 (20dB) to bring the signal back to the original level. The noise is now 10 times as intense. This appearance is often described as "grainy", "snowy", or "noisy", or "low SNR" (Signal to Noise Ratio). Note that it is essentially patternless. The other artifacts shown here are NOT, and should not be referred to as "noise". Raw data here. Back to Reference Image or Top RF Inhomogeneity
A 2-inch square piece of copper foil was placed at the left side of the phantom to intercept the B1 field. The shadowing (due to RF eddy currents in the copper foil) results in almost complete signal dropout. No change in the apparent pulse calibration was observed. Similar effects - uneven illumination - occur with some types of RF coils, particularly, but not limited to, surface coils . Back to Reference Image or Top Shimming
The effect of mis-set shim currents (i.e. Bo inhomogeneity) depends on the imaging technique. These two examples were acquired with the ZXY shim current misadjusted. The left image above was acquired with the spin-echo sequence used for most of the rest of the gallery, and a 1.86 A ZXY error. The only sign of the shim problem is a tiny geometric distortion - less than 2 pixels (viewed along the Z axis). The extent of the distortion depends on the readout gradient strength, of course. The center image shows the difference between the distorted and the reference image. The right image was acquired with a gradient echo sequence, using similar excitation parameters, but only one third of the current error (0.61 A). The Z (slice) component of the shim error now causes a tremendous signal loss (failure to refocus) in areas where it is large. Note the presence of receiver turn-on artifact in the gradient echo image (see also: FID artifact). Back to Reference Image or Top Stuck Bit
Once again, viewit has been used to modify the data, by performing a bitwise OR with a binary value having one bit set to zero (e.g. 1111011111). This artifact is uncommon with modern hardware, but not impossible. It can occur if there are bad memory locations or bad connectors in the parallel data bus. Notice the similarity to the Digitizer Quantization example. The severity depends on which bit is stuck. Back to Reference Image or Top Susceptibility
A US quarter-dollar coin (1992) was placed next to the phantom. The metallic object is too small, or too poorly conductive, to cause significant RF inhomogeneity (compare with the real thing, above). But it does distort the magnetic field, in a manner similar to, but more severe than, a shimming error. Careful inspection reveals a small distortion of the circular phantom profile at the left side. A characteristic feature of this type of distortion is the accompanying bright-spot/dark-spot effect. In this case, the edge of the phantom near left center is a bit darker, and there is a brighter area just below it. This differs from the RF inhomogeneity effect in that signal appears to be "conserved". The effect is subtle in this image because the coin is distant from the signal volume. Closer contact can produce much more severe effects. Back to Reference Image or Top Experimental DetailsAll data were acquired on the Beckman MRI system, using the SMIS volume RF coil. Unless noted, the object is a phantom consisting of a 4oz (125ml) Nalgene plastic bottle with 5mM CuSO4 in water. T1 ~ 300 msec, T2 ~ 270 msec. Unless noted, all images are acquired with a very conventional 2D spin echo spin-warp imaging scheme. Unless noted, data were processed using a simple two-dimensional Fourier transform (without time-domain weighting), and are presented as the magnitude of the signal, scaled by a fixed factor of 0.01. The parameters for the standard (Reference) image follow: (from rfi.mrd) C:\BMRL\SEQLIB\SE2D.PPL OBSERVE_FREQUENCY 170.029470 MHz rec_freq, 0 tr, 400 msec te, 20 msec sample_period, 20.0 µsec (50 kHz spectral width) no_samples, 256 no_views, 256 no_averages, 1 view_block, 1 slice_block, 1 phase_cycle, 1 orientation, 1 gr_var, -512 gp_init_var, -512 :FOV 75.0003 slice_off_freq, 0 :SLICE_THICKNESS gs_var, -128, 10 :SLICE_SEPARATION slice_offset, 150, 1 :NO_SLICES no_slices, 1 :SLICE_INTERLEAVE slice_interleave, 1 flip_angle, 90 obs_power, 4 gr_comp_scale, 0 gs_comp_scale, 0 tc1, 1000 tc2, 5000 g_spoil, 820 triggered, 1 no_dummy_scans, 0 :DISCARD no_discard, 2 :DATA_TYPE 0x13 :_ObserveTransmitGain -33 :_ObserveReceiverGain 250 :_ObserveReceiverAttenuation -170 Back to Reference Image or Top
C.D.Gregory: April 1, 1997
|
|
