 |
 |
Hi Helmut, Thank you so much for your great help! I will use your segmented GM for now and continue trying to figure out why it fails at my side. Thanks a lot!! MA On Sun, May 10, 2015 at 12:54 AM, H. Nebl <[log in to unmask]> wrote: > Dear Marris, > > Well, I'm not really sure what this screenshot shows (the segmented grey > matter image?), and how it was visualized. It looks like a binary image > with zeros and ones, but maybe this is just due to some display settings, > coding anything non-zero as white. See attachements for 1) the raw > structure and the c1 grey matter image obtained with "Segment" in SPM12, > visualized within SPM, and 2) the grey matter tissue probability map as > provided with SPM and the wc1 normalized grey matter image as obtained with > the same routine. > > The output seems to be reasonable. It is not perfect, e.g. some white > matter is classified as GM (see raw&c1_WML). This is not surprising though, > as the raw image clearly shows some intensity changes, pointing to > periventricular white matter lesions (and/or artefacts). It's not a matter > of algorithm (SPM12 Segment vs. SPM8 vs. VBM8), but would require a special > preprocessing pipeline to avoid misclassifications (e.g. some additional > tissue probabiliy maps coding regions where WML are likely to occur). > > What bothers me a little is the contrast of the raw image. When looking at > e.g. the topmost slices or at some of the sulci there is hardly any > contrast between grey matter and white matter. It would thus be impossible > to properly segment the image manually. The segmentation algorithm results > in a reasonable image, but this is also because it is forced to classify > things as grey matter due to spatial constraints. Thus it would be > interesting to see the true raw image. I have doubts the current one is > really the true one. When displaying it with a different intensity mapping > it becomes evident that voxels around the head are coded as zero = there is > no noise, which is very surprising given the usual low-intensity noise. > Based on the sharp edges some correction must have been performed on that > image. And then the question arises, which type of correction was performed > and how did the raw data look like before that step? It might just have > been some bias correction, but it might also have been prospective motion > corrections like PROMO. > > For illustrative purpose I've also attached an unaltered T1 image > screenshot affected by motion (motion artefacts can be much worse). > Posteriorly, at the enlarged sulci, it's (almost) impossible to identify > the cortical sheet, it looks like CSF -> brain tissue, not CSF -> GM -> WM. > Anteriorly, quite some smearing is evident. Maybe this was also the case in > your file, after some motion correction the blurring might have been > reduced, but the contrast might still be quite low. This is just guessing > of course. It would be necessary to see some of the other images to get an > impression whether this is due to the scanner/sequence settings or whether > it's a problem in particular volumes. The contrast might have been low due > to the settings, but it might also have been due to motion or an artefact > introduced during the correction step. > > Best > > Helmut >