Motion Artefact Propagation
View the random motion artefacts page.
It's the phase encoding direction. You can tell because the artefact that propagates across the whole image is in this direction. Look at these two images:
[movement in PE direction]
[movement in FE direction]
The (simulated) movement occurred during the acquisition of the 5 lines just above the very middle of k-space. For this reason, no matter which way the movement occurred, there is a large imprint of the phantom displaced in the direction of movement. But notice this is simply the correct representation of where the phantom was during the acquisition of those lines. However, the ghosting effect due to the FFT reconstruction, which propagates across the whole image, is in the phase encoding direction in both situations (up-down in this case).
The large ghost due to the fact that the phantom was moved during the acquisitions of the central lines of k-space would be much less pronounced if the movement occurred during the acquisition of the peripheral lines of k-space. Also, in "real life" movement in either direction is unlikely to be as exaggerated as this, and so there probably wouldn't be a large imprint, so much as a little blurring. The ghosting effect would still be present across the whole image, however. But then again, the grey-scale windowing would probably omit a lot of that too!
See also this image artefacts question.
Further reading on this topic:
Books: MRI From Picture to Proton p79