Acquisition (artifacts)
- SNR - signal to noise - single image (contrast to noise, dropout/susceptibility; gray/white matter contrast in structurals)
- SNR - temporal (e.g., maps of mean / std. deviation across time)
- ghosting - image wrap-around artifact*
- image intensity - high enough to avoid information loss? (e.g., >1000)
- distortions - spatial distortions in acquisition
Motion artifacts mainly propagate in the phase-encode direction. This is due to movement of the spins between 2 excitations or between phase-encoding and signal reading: in the first case, the spins will not be recorded at the same position between excitations, in the second case, their phase-encoding will not be correct. As a result, the phase-encoding of these voxels is corrupted and this will be responsible for artifacts in the phase-encode direction.
On the other hand, signal sampling and spatial-encoding in the frequency-encode direction are done so fast that physiological motion will only produce a small amount of spatial blurring in that direction.
- Spikes - gradient artifacts and bad images *
- drift - large signal drift over time (look at FFT or timeseries)
- periodic noise - low-frequency noise artifacts
When the movements are periodical (cardiac beats, arterial or CSF pulsations, respiration), they can produce ghost images, propagated in the phase-encode direction, even outside the anatomic limits. The intensity of these ghost images becomes more extreme with the intensity of the moving structure and with the amplitude of movement. These ghost images can show up as an increase or decrease of the true image signal.
The spacing between ghost images varies with the direction of the movement, its amplitude and its periodicity relative to the phase-sampling interval (TR).
- streaks/striping in images (RF room leaks)
Processing
- orientation (reconstruction/header problems)
- skull stripping or segmentation failures (missing brain)
- coregistration (anatomical to functional overlap)
- normalization (warping was ok? bad alignment/distortion?)
- movement estimates (reasonable? too much movement?)
Modeling/statistics
- colinearity and predictor variance in model
- mask (which voxels are analyzed?)
- image registration (consistent across all images in analysis?)
- contrast scaling (consistent across all images?)
- indicators of task-correlated artifacts: global shift, large std across contrast
- outliers (very unusual images)
Reference: http://wagerlab.colorado.edu/wiki/
http://www.imaios.com/en/e-Courses/e-MRI/Image-quality-and-artifacts/image-quality
