Dear all ! I know, this has been discussed a lot in the forum, however, I am still not sure, whether my analysis pipeline is correct. I simply want to compare volumes of FS vs. manual segmentation in the hippocampus and calculate the dice coefficient. So these are my steps:
1. mri_convert T1.mgz T1.nii 2. Register (e.g. with FLIRT) T1.nii to original_mprage.nii (where the manual segmentation was done), output: registration.matrix 3. mri_extract_label aseg.mgz 17 left_FreeSurfer_HC.nii mri_extract_label 53 right_FreeSurfer_HC.nii 4. Register left_ & right_FreeSurfer_HC.nii to original_mprage.nii by using registration.matrix 5. mri_compute_overlap left_Freesurfer_ HC left_manual_HC
Now here are my questions: A) I've read that I should use mri_label2vol instead of mri_convert Is that correct and how would I process that afterwards ?
B) The volumes form aseg.mgz are different (in my case: bigger) to the aseg.stats because of partial volume correction. Can I use partial volume corrected hippocampal masks ?
C) Where can I find information about the segmentation method of the hippocampus, used in FreeSurfer, especially regarding the HC borders ? Because otherwise it might be difficult to calculate and interpret dice coefficients.
Best regards and thanks for any help, Bernd
A) I don't think this matter, but Doug can correct me if I'm wrong. B) You can't partial-volume correct the DICE, but you can compare the volumes after partial-volume corrections. C) There are references in our Neuron paper on the definition of the hippocampus (Caviness, Makris, Kennedy papers)
cheers Bruce
On Tue, 11 Nov 2014, Bernd Merkel wrote:
Dear all ! I know, this has been discussed a lot in the forum, however, I am still not sure, whether my analysis pipeline is correct. I simply want to compare volumes of FS vs. manual segmentation in the hippocampus and calculate the dice coefficient. So these are my steps:
- mri_convert T1.mgz T1.nii
- Register (e.g. with FLIRT) T1.nii to original_mprage.nii (where the
manual segmentation was done), output: registration.matrix 3. mri_extract_label aseg.mgz 17 left_FreeSurfer_HC.nii mri_extract_label 53 right_FreeSurfer_HC.nii 4. Register left_ & right_FreeSurfer_HC.nii to original_mprage.nii by using registration.matrix 5. mri_compute_overlap left_Freesurfer_ HC left_manual_HC
Now here are my questions: A) I've read that I should use mri_label2vol instead of mri_convert Is that correct and how would I process that afterwards ?
B) The volumes form aseg.mgz are different (in my case: bigger) to the aseg.stats because of partial volume correction. Can I use partial volume corrected hippocampal masks ?
C) Where can I find information about the segmentation method of the hippocampus, used in FreeSurfer, especially regarding the HC borders ? Because otherwise it might be difficult to calculate and interpret dice coefficients.
Best regards and thanks for any help, Bernd
I would use mri_label2vol to map your manual labels into the conformed space (using mri_label2vol) (and assuming that your native space has voxel sizes 1mm or greater). This is just a single step (I did not quite follow all the steps you have below). No need to run any registration, just follow the instructions here http://surfer.nmr.mgh.harvard.edu/fswiki/FsAnat-to-NativeAnat to get the proper registration file.
doug
On 11/11/14 10:12 AM, Bruce Fischl wrote:
A) I don't think this matter, but Doug can correct me if I'm wrong.
mri_label2vol will do better partial volume correction
B) You can't partial-volume correct the DICE, but you can compare the volumes after partial-volume corrections. C) There are references in our Neuron paper on the definition of the hippocampus (Caviness, Makris, Kennedy papers)
cheers Bruce
On Tue, 11 Nov 2014, Bernd Merkel wrote:
Dear all ! I know, this has been discussed a lot in the forum, however, I am still not sure, whether my analysis pipeline is correct. I simply want to compare volumes of FS vs. manual segmentation in the hippocampus and calculate the dice coefficient. So these are my steps:
- mri_convert T1.mgz T1.nii
- Register (e.g. with FLIRT) T1.nii to original_mprage.nii (where the
manual segmentation was done), output: registration.matrix 3. mri_extract_label aseg.mgz 17 left_FreeSurfer_HC.nii mri_extract_label 53 right_FreeSurfer_HC.nii 4. Register left_ & right_FreeSurfer_HC.nii to original_mprage.nii by using registration.matrix 5. mri_compute_overlap left_Freesurfer_ HC left_manual_HC
Now here are my questions: A) I've read that I should use mri_label2vol instead of mri_convert Is that correct and how would I process that afterwards ?
B) The volumes form aseg.mgz are different (in my case: bigger) to the aseg.stats because of partial volume correction. Can I use partial volume corrected hippocampal masks ?
C) Where can I find information about the segmentation method of the hippocampus, used in FreeSurfer, especially regarding the HC borders ? Because otherwise it might be difficult to calculate and interpret dice coefficients.
Best regards and thanks for any help, Bernd
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Bernd Merkel -
Bruce Fischl -
Douglas Greve