Griselda and others,
>
>Dear SPMers,
>
>Has anyone developed a method to automatically
>detect (and replace) axial slices with background noise?
>We observed that the median (and also the mode)
>of the slice is significantly greater than
>the median of adjacent slices or the same slice
>in adjacent volumes:
>http://200.206.142.103/pbfm/pessoas/griselda/web/BEM/BEM_BCK_RF.html
>
>Many of our axially acquired slices show one vertical line
>of RF artefact, always in the same position, forming one
>noisy sagittal slice. Should we replace it with the mean of
>the adjacent sagittal slices in the same volume?
I'd like to know what other people are doing as well. I've
spent some time on identification of artifacts in our time
series, during the last year. However, I've been hesitant to
use an automated approach because one runs the risk of
throwing out good images if the identification scheme isn't
finely tuned. In addition, different types of artifacts
appear differently using different tests (e.g., global
unthresholded mean, global modes, realignment parameter
examination, etc...).
We tend to use a tree-like approach, searching all scans with
simple, quick tests before performing more time-intensive ones
on noisy candidates. Using unthresholded mean image intensity
values for a time series usually indicates artifacts in our
spiral images as a series of spikes. We've found that for
non-motion-related artifacts, looking at intensities in
regions of air is more useful (because it's all supposed to be
black, all the time!). If we find spikes, we can look at air
and head mean intensity values per slice, to isolate
potentially goofy slices or image volumes. Noting these, I
feed a small number of allegedly goofy images along with
allegedly normal images into a movie viewer tool. Looping
quickly through a small number of these images will show
artifacts fairly well, perhaps with a little contrast
enhancement. You can alternatively look at a standard
deviation image for the time series.
The tree process helps us to zoom in on potential artifacts,
but leaves the final evaluation process to human judgement.
Other ideas out there?
Nathan
Dr. Nathan Yanasak
Program Coordinator
UGA Human Neuroimaging Facility
Department of Psychology
University of Georgia
&
Research Statistician
Department of Radiology
Medical College of Georgia
(706) 542-7601 my desk (Rm 416 Psych, UGA)
(706) 339-7250 cell
(706) 542-3076 Rm 415 neuropsych lab
(706) 542-3275 fax
(706) 721-3602 MCG number
email: [log in to unmask]
|