HI Anthony
I would make sure to include an entire brain labeling in the training set. You can do this by pasting your manual labels into our aseg of the subjects in question.
cheers Bruce
On Fri, 20 Apr 2012, Anthony Dick wrote:
Hello,
I developed a new atlas for parcellating the cerebellum, based on 20 brains manually segmented by Jörn Diedrichsen, which he was kind enough to share (note these also formed the basis for his 2009 paper in Neuroimage (46; 39-46).
I got the segmentation to work, but I would not say it works well. As you can see from the attached image, it is not ready for prime-time. On the left is the brain of subject 1, manually segmented. On the right is the brain of the same subject with the automated sementation, based on the developed atlas.
My question is, how can I improve it? I realize this is an open question and potentially difficult to answer without some probing, but this is a resource I think a lot of users would like to have access to, so perhaps it is worth the effort.
Anthony