An image artifact is any feature which appears in an image which is not present in the original imaged object. An image artifact is sometime the result of improper operation of the imager, and other times a consequence of natural processes or properties of the human body. Artifacts are typically classified as to their source. The following table summarizes a few of these.
| Artifact | Cause |
|---|---|
| RF Quadrature | Failure of the RF detection circuitry |
| Bo Inhomogeneity | Metal object distorting the Bo field |
| Gradient | Failure in a magnetic field gradient |
| RF Inhomogeneity | Failure of RF coil |
| Motion | Movement of the imaged object during the sequence |
| Flow | Movement of body fluids during the sequence |
| Chemical Shift | Large Bo and chemical shift difference between tissues |
| Partial Volume | Large voxel size |
| Wrap Around | Improperly chosen field of view |
An example of each of the artifacts is presented next. The reader is cautioned that a problem with the imager can manifest itself in a number of ways. Therefore not all artifacts of a given type will appear the same.
RF quadrate artifacts are caused by problems with the RF detection circuitry. More specifically, these problems are typically associated with what was referred to in the hardware section as the quadrature detector. These problems arise from improper operation of the two channels of the detector. For example if one of the amplifiers has a DC offset in its output the Fourier transformed data can display a bright spot in the center of the image. If one channel of the detector has a higher gain than the other it will result in a ghosting of objects diagonally in the image. This artifact is the result of a hardware failure and must be addressed by a service representative.
All magnetic resonance imaging assumes a homogeneous Bo magnetic field. An inhomogeneous Bo magnetic field will cause distorted images. The distortions can be either spatial, intensity, or both. Intensity distortions result from the field homogeneity in a location being greater or less than that in the rest of the imaged object. The T2* in this region is different, and therefore the signal will tend to be different. For example, if the homogeneity is less, the T2* will be smaller and the signal will be less. Spatial distortion results from long-range field gradients in Bo which are constant. They cause spins to resonate at Larmor frequencies other than that prescribed by an imaging sequence.
The animation window contains an image of four water filled straight tubes positioned so as to form a square. The magnetic image shows a severe bending in one of the tubes due to a nonuniformity in the Bo magnetic field.
Artifacts arising from problems with the gradient system are sometimes very similar to those described as Bo inhomogeneities. An gradient which is not constant with respect to the gradient direction will distort an image. This is typically only possible if a gradient coil has been damaged. Other gradient related artifacts are due to abnormal currents passing through the gradient coils. In this image the frequency encoding (left/right encoding) gradient is operating at half of its expected value.
An RF inhomogeneity problem is a variation in intensity across an image. The cause is either a nonuniform B1 field or an nonuniform sensitivity in a receive only coil. Some RF coils, such as surface coils, naturally have variations in sensitivity and will always display this artifact. The presence of this artifact in other coils represents the failure of an element in the RF coil or the presence of nonferromagnetic material in the imaged object. For example a metal object which prevents the RF field from passing into a tissue will cause a signal void in an image.
The accompanying sagittal image of the head contains an RF inhomogeneity artifact in the region of the mouth.
(See arrow.) The patient has a large amount of non ferromagnetic metal dental work in the mouth. The metal shielded the regions near the mouth from the RF pulses thus producing a signal void. The Dental work did not significantly distort the static magnetic field Bo.
As the name implies, motion artifacts are caused by motion of the imaged object or a part of the imaged object during the imaging sequence. The motion of the entire object during the imaging sequence generally results in a blurring of the entire image with ghost images in the phase encoding direction. Movement of a small portion of the imaged object results in a blurring of that small portion of the object across the image.
To understand this artifact picture the following simple example. A single small spin containing object is imaged.
The central portion of the MX
raw data will look something like this.
The frequency of the waves will be related to the position in the frequency encoding direction and the variation in phase of the waves will be related to the position in the phase encoding direction. Fourier transforming first in the frequency encoding direction yields a single oscillating peak.
Viewing the data as a function of a phase shows this more clearly.
Fourier transforming last in the phase encoding direction yields a single peak at the location of the original object.
Now picture the same example except that midway through the acquisition of phase encoding steps the object moves to a new location in the frequency encoding direction.
The central part of the MX raw data looks like this.
Fourier transforming first in the frequency direction gives two oscillating peaks which abruptly stop oscillating.
Viewing the data as a function of a phase shows this more clearly.
Fourier transforming in the phase encoding direction gives several repeating peaks at the two frequencies. This is because the Fourier pair of an abruptly truncated sine wave is a sinc function. The magnitude representation of the data makes all the peaks positive.
The solution to a motion artifact is to immobilize the patient or imaged object. Often times the motion is caused by the heart beating or the patient breathing. Both of which can not legally be eliminated. The solution in these cases is to gate the imaging sequence to the cardiac or respiratory cycle of the patient. For example if the motion is caused by pulsing artery, one could trigger the acquisition of phase encoding steps to occur at a fixed delay time after the R-wave in the cardiac cycle. By doing this the artery is always in the same position.
Similar gating could be done to the respiratory cycle. A disadvantage of this technique is that the choice of TR is often determined by the heart rate or respiration rate. Imaging techniques designed to remove motion artifacts are given different names by the various manufacturers of magnetic resonance imagers. For example, a few names of sequences designed to remove respiratory motion artifacts are respiratory gating, respiratory compensation, and respiratory triggering.
The accompanying axial image of the head shows a motion artifact.
A blood vessel in the posterior side of the head moved in a pulsating motion during the acquisition. This motion caused a ghosting across the image.
Flow artifacts are caused by flowing blood or fluids in the body. A liquid flowing through a slice can experience an RF pulse and then flow out of the slice by the time the signal is recorded. Picture the following example. We are using a spin-echo sequence to image a slice. Here the timing diagram and side view of the slice are shown.
During the slice selective 90o pulse blood in the slice is rotated by 90o. Before the 180o pulse can be applied, the blood which experienced the 90o pulse has flown out of the slice.
The slice selective 180o pulse rotates spins in the slice by 180o.
However the blood in the slice has its magnetization along +Z before the pulse and along -Z after the pulse. It therefore yields no signal. By the time the echo is recorded the slice has only blood in it which has not experienced the 90o
or the 180o pulse. The result is that the blood vessel which we know to contain a high concentration of hydrogen nuclei yields no signal.
Here is an example from an axial slice through the legs. Notice that the blood vessels appear black even though they contain a large amount of water.
In a multislice sequence, the slices could be positioned such that blood experiencing a 90o pulse in one slice can flow into another slice and experience a 180o rotation and into a third and contribute to the echo. In this case the vessel will have a high signal intensity. The effect is usually that some slices have low signal intensity blood vessels and others have high signal blood vessels.
A chemical shift artifact is caused by the difference in chemical shift (Larmor frequency) of fat and water. The artifact manifests itself as a misregistration between the fat and water pixels in an image.
The difference in chemical shift is approximately 3.5 ppm which at 1.5 Tesla corresponds to a frequency difference between that of fat and water is approximately 220 Hz. During the slice selection process there is a slight offset between the location of the fat and water spins which have been rotated by an RF pulse. This difference is exaggerated in this animation.
During the phase encoding gradient the fat and water spins acquire phase at different rates. The effect being that fat and water spins in the same voxel are encoded as being located in different voxels. In this example all nine voxels have a red water vector. The center voxel has some fat magnetization in addition to the water.
In a uniform magnetic field the vectors precess at their own Larmor frequency.
When a gradient in the magnetic field is applied, such as the phase encoding gradient, spins at different x positions precess at a frequency dependent on their Larmore frequency and field. In this example the fat vector has the same frequency as the water vector in the voxel to its right.
When the phase encoding gradient is turned off each vector has acquired a unique phase dependent on its x position.
During the frequency encoding gradient, fat and water spins located in the same voxel precess at rates differing by 3.5 ppm. The net effect is that the fat and water located in the same voxel are encoded as being located in different voxels. In this example the fat vector in the center voxel possesses a phase and precessional frequency as if it was located in the upper right voxel.
The resultant image places the fat in the upper right voxel rather than in the center.
The magnitude of the effect is proportional on the magnitude of the Bo field and inversely proportional to the sampling rate in the frequency encoding direction. For a constant sampling rate, the larger Bo, the greater the effect. At 1.5 T and a 16 kHz sampling rate, the effect is approximately 3.5 pixels. At 0.5 T and a 16 kHz sampling rate, the effect is approximately one pixel. In this axial slice image through the legs there is a chemical shift artifact between the fat and muscle in the legs.
A partial volume artifact is any artifact which is caused by the size of the image voxel. For example, if a small voxel contains only fat or water signal, and a larger voxel might contain a combination of the two, the large voxel possess a signal intensity equal to the weighted average of the quantity of water and fat present in the voxel. Another manifestation of this type of artifact is a loss of resolution caused by multiple features present in the image voxel.
Here is a comparison of two axial slices through the same location of the head. One is taken with a 3 mm slice thickness and the other with a 10 mm thickness. Notice the loss of resolution in the 10 mm Thk image.
The solution to a partial volume artifact is a smaller voxel, however this may result in poorer signal-to-noise ratios in the image.
A wrap around artifact is the occurrence of a part of the imaged anatomy, which is located outside of the field of view, inside of the field of view. This artifact is caused by the selected field of view being smaller than the size of the imaged object. Or more specifically the digitization rate is less than the range of frequencies in the FID or echo. The solution to a wrap around artifact is to choose a larger field of view, adjust the position of the image center, or select an imaging coil which will not excite or detect spins from tissues outside of the desired field of view.
In the accompanying sagittal image of a breast, the portion of the image below the arrow should appear on the top of the image.
This portion was located at a position that had a greater resonance frequency than the digitization rate. As a consequence, it was wrapped
around and appears at the bottom end of the image.
Many newer imagers employ a combination of oversampling, digital filtering, and decimation to eliminate the wrap around artifact. Oversampling creates a larger FOV, but generates too much data to be conveniently stored. Digital filtering eliminates the high frequency components from the data, and decimation reduces the size of the data set. The following flowchart summarizes the effects of the three steps by showing the result of performing an FT after each step.
Let's examine oversampling, digital filtering, and decimation in more detail to see how this combination of steps can be used to reduce the wrap around problem.
Oversampling is the digitization of a time domain signal at a frequency much greater than necessary to record the desired field of view. For example, if the sampling frequency, fs, is increased by a factor of 10, the field of view will be 10 times greater, thus eliminating wrap around. Unfortunately digitizing at 10 times the speed also increases the amount of raw data by a factor of 10, thus increasing storage requirements and processing time.
Filtering is the removal of a select band of frequencies from a signal. For an example of filtering, consider the following frequency domain signal.
Frequencies above fo could be removed from this frequency domain signal by multipling the signal by this rectangular function.
In MRI, this step would be equivalent to taking a large FOV image and setting to zero intensity those pixels greater than some distance from the isocenter.
Digital filtering is the removal of these frequencies using the time domain signal. Recall from
Chapter 5
that if two functions are multiplied in one domain (i.e. frequency), we must convolve the FT of the two functions together in the other domain (i.e. time). To filter out frequencies above fo from the time domain signal it must be convolved with the Fourier transform of the rectangular function, a sinc function.
(See Chapter 5.)
This process eliminates frequencies greater than fo from the time domain signal. Fourier transforming the resultant time domain signal yields a frequency domain signal without the higher frequencies. In MRI, this step will remove image components
fo / 2 g Gf
away from the center of the image.
Decimation is the elimination of data points from a data set. A decimation ratio of 4/5 means that 4 out of every 5 data points are deleted, or every fifth data point is saved. Decimating the digitally filtered data above, followed by a Fourier transform, will reduce the data set by a factor of five.
High speed digitizers, capable of digitizing at 2 MHz, and dedicated high speed integrated circuits, capable of performing the convolution on the time domain data as it is being recorded, are used to realize this procedure.
Gibbs ringing is a series of lines parallel to a sharp intensity edge in an image. The ringing is caused by incomplete digitization of the echo. This means the signal has not decayed to zero by the end of the acquisition window, and the echo is not fully digitized. (The reader is encouraged to prove this using the convolution theorem.) This artifact is seen in images when a small acquisition matrix is used. Therefore, the artifact is more pronounced in the 128 point dimension of a 512x128 acquisition matrix.
In the following example, a rectangular object with a spatially uniform signal is imaged. An inadequate number of points are collected in the horizontal (x) direction. The resultant image displays a ringing in the intensity at the edge. The animation window displays the upper right hand corner of this image and a plot of signal intensity.
Copyright © 1996-99 J.P. Hornak.
All Rights Reserved.