MR Images are loaded with amusing artifacts. These images give some typical examples.
This is a good quality (3T) image of a phantom with a 256x256 matrix
Small motions of the phantom can result in severe artifacts. The nature of the artifact depends on the timing of the motion with respect to the acquisition. Here the phantom was moved during the middle of the sequence, resulting in low spatial frequency (low-k value) artifacts.
If the motion occurs during collection of the high spatial freuqencies (usually at the beginning and end of the acquisition period, the artifacts are very different:
When the magnetization of the phantom is not uniform (due to the presence of ferromagnetic materials, inaccurate shimming, other foreign bodies), the images will be distorted. The degree of distoration depends on the "bandwidth" of the acquisition (lower bandwidths -> more distortion). The image below was collected with a low bandwidth (2 kHz) with the shim supplies turned off. The distortions are obvious.
Aliasing is the term given to the artifact that occurs when features outside of the FOV of the image are folded back in. This happens when the MR smapling rate is lower than twice the maximum frequency of the image features. The out of field data are represented at the wrong frequency, and after Fourier transformation, end up inside of the image field of view.
In the event that the prescan did not complete properly, so that the receiver is overloaded, the data will be clipped in time. After Fourier transformation, the artifacts is more sinc-like. Because the signal is largest near the center of k-space, the artifacts are typically of low spatial frequency. Often the low spatial frequency features are lost.
Because the FT is infinite, yet only a limited signal is collected in MRI, there is a complex attenuation and distortion of the signal from small image features. Basically, the signal from tiny features "rings" into neighboring pixels, resulting in loss of signal in the actual location and artifactual signals elsewhere in the image.
Suppose that you have a single intense pixel. The Fourier transform of that data point, which is the raw MRI time data, will be a broad sinc (sin(t)/t) function. Since the sinc function has infinite tails (in time, now) it is not possible to capture all of it. Instead, we captuer only the middle portion.
To convert the time data back to an image, it is Fourier transformed, again. The pictures below show a single bright pixel image, before and after Fourier transform MR imaging. The stack plots, below, show the intensities in the area outlined in light gray (the outline is not part of the original image...)
Below is a brain shaped "phantom" with a single intense pixel. The line plot shows the signal intensities along a horizontal cut through the bright feature.
A surface plot shows the clearly focused region of greatly increased signal.
After Fourier transformation, the images show a radical signal change. On the left is an image acquired with a 64x64 matrix, and to the right with a 32x32 matrix. The lower resolution image shows tremendous artifact. Note also that the edges of the phantom ring. This is because they too contain high spatial frequencies (like the bright spot). In clinical images such edge ringing can be mistaken for things such as a spinal syrinx.
Were this an "activation" in fMRI, it is clear that the lower resolution images can result in not only loss of signal, but in displacement of the apparent locus of activation and pseudo activations many pixels away.
The surface plots for a 64x64 image (left) and a 32x32 image (right) show the dramatically reduced and distorted effects of the small area of "activation" after MR imaging.
The next series of pictures demonstrates the same effects in MR head images. To the left, the basic image (without any extra bright pixels). To the right, the same image with a bright focus (arrow). added.
![[BMC Home Page]](../../BMC_IMAGES/BMCbutton96.GIF)