Hi everyone,
We are trying to segment images from the ADNI database and we realized that various of the segmentations performed with recent versions of CAT12 presented some issues:
1) In several segmentations, some fragments of the skull are considered as part of the GM tissue. Find linked below two examples of how the "rp1" image of two different subjects looks like when segmented with the latest version of CAT12. We also observed these segmentation issues in the release 1318.
Subject 1: https://i.imgur.com/46xH3Kg.png
Subject 2: https://i.imgur.com/ZAIy7CH.png
And these are the same images segmented with a previous CAT12 release (r1158):
Subject 1: https://i.imgur.com/EcmSUx7.png
Subject 2: https://i.imgur.com/a2pSMWQ.png
We noted that, by default, the skull-stripping parameter is set to "medium (0.5)" in the r1158, whereas in the latest version it is set to 0. Setting this parameter to 0.5 in the latest CAT12 version appears to partly solve this problem, although there are still some images that are not correctly segmented in the latest versions whatever the value of this parameter is.
2) There appears to be an "oversegmentation" of the CSF in other cases, such as this subject segmented with the latest version of CAT12 (https://i.imgur.com/XZsIqdF.png, CSF volume: 965 cm^3) as compared with the same subject segmented with the release 1158 (https://i.imgur.com/F1X9DxB.png, CSF volume: 470 cm^3).
3) We also found several cases in which large fragments of WM tissue are considered to be GM tissue (e.g., https://i.imgur.com/kzq7QC5.png). This happens in all the 3 CAT12 versions that we have tested, so it is likely an issue related with the quality of the image and its bad contrast between GM and WM tissues. We tried changing various parameters but we did not observe any improvement. Do you have any advice on this?
Thank you.
|