I have an issue regarding the correction for CSF fraction. I did follow all the instructions of the concensus paper by J Near and colleagues in NMR and biomedicine and many other papers. Here is the issue> we compared the hypothalamus of 2 cohorts of healthy subjects at 2 different periods of the year at 3T with a STEAM sequence provided by CMRR. I used GANNET and SPM12 to perform the segmentation. For the first period of the year the CSF fraction is 18% and for the second period it is …74%. When I use all recommendations of the concensus paper, the metabolite concentrations are largely overestimated. One of our reviewers speculated that the reason for this is that the acquisition voxel was misplaced during the second period compared to the first one. But I verified this over and over and it is not the case… The presence of the third ventricle within the voxel might be the reason. Can anybody help me ? And anyway what would be expected fom the quantification in case of a too large CSF fraction compared with GM and WM?
Thanks a lot!
Have you looked at the segmentation maps to see if there was some kind of error in the segmentation itself?
Thanks a lot for your reply. Indeed, I have looked at the segmentation maps. For a few datasets , it seems that the segmentation did not work at all giving 100 % CSF but for others we had 47% GM and 52% CSF and the segmentation looked fine. This only happened for the second period and second measurement. The sheep were healthy, we only expected a few physiological and hormonal changes. We resorted to use the segmentation of the first period but obviously this is not the right thing to do…
Hold on - is this a typo or did you really scan sheep?!
Nope, no typo… Female sheep…But I can\t understand what is going on …
But the brain is very similar, but half-size, to the human brain…
I’m pretty sure that you can’t segment sheep brain images based on human tissue probability maps.
Could this help? I didn’t look too closely but they seem to have a sheep tissue probability map in there.
Thanks a lot for this. Good idea but we have a sheep brain atlas and obtained the tissue probability maps, which were used for the segmentation…
Thanks a lot, Georg for sending the reference, which we are aware of, but we have our own sheep brain atlas, which was improved compared to the reference you sent and we have the probability maps calculated for our animals during the 2 periods of the year.
Ella et al. J Comp Neurol. 2017 Feb 15;525(3):676-692.
@njust, so you do have the tiissue probability maps (TPM’s) for sheep brain? But did you make sure to modify Gannet to select these maps? Gannet uses the default TPM’s that SPM uses (based on the MNI template). You can modify the file selection in
CallSPM12segmentation.m. Also, it’s important to note that the parameters set for segmentation are the default ones. They may not be optimal for sheep brain.
We changed all the necessary files and tested various things. The outcome was not changed.
Thanks a lot to MRSHub and especially Georg and Mark for their help. After checking and re-checking, I think that the problem probably comes from a mis-orientation of the scans compared to the probability maps…