Jamie,
There should be some other information in the log informing why the label is 'bad'. There are checks in the mri_ca_train for labels having strange coordinates, mainly there to catch things like a right- hippocampus label in the left hemisphere. So look for those type error messages in the mri_ca_train output.
Nick
On Tue, 2009-04-28 at 15:47 -0500, Jamie Hanson wrote:
Hi Jamie, what do you mean by "bad labels"?
when going through each subject in mri_ca_train, some subjects output the following: ERROR: mri_ca_train: possible bad training data! subject: /study/eemri/data/proc/freesurfer/30_aseg
and then after all the subjects are finished in mri_ca_train, it says: ERROR: mri_ca_train check found 6 subjects with bad labels!
i edited everything by hand (and had another raters also check the edits) and there is nothing visually, i can see that is wrong w/ the segmentation edits. so i wasn't sure if something else went wrong and i should remove those 6 subjects?
best, jamie.
Bruce On Tue, 28 Apr 2009, Jamie Hanson wrote:
Hi Freesurfer folks-
I am in the midst of creating a custom aseg atlas and had a quick question regarding when to use v. discard subjects.
I just ran the second iterations of mri_ca_train (when you use all the subjects in subjects.csh and are training from segmented subjects using M3D_ONE) and I have 6 (of 18) subjects "with bad labels". Is it best to discard those subjects and just use the successfully labeled subjects? I wasn't sure if only included those good subjects would perhaps boost my aseg atlas accuracy.
Thanks much.
Best, jamie.