Re: [Freesurfer] V6beta: recon-all --noasegfusion does not initialize using longbase aseg.mgz; how to handle aseg edits in longitudinal stream
Dear Martin,
one more thing: When the error is large and present in both timepoints (as is my case), I think that necessity of editing the same error in both timepoints can easily lead to error (maybe even demonstrated as false positive effect) caused by non-systematic editing. When editing comprises many voxels one cannot realistically assure the perfect consistence in editing.
Editing only once (in base) and initialize mri_ca_label in long from base can avoid this problem. So I think the initialization of mri_ca_label from base is not a bad option for such cases.
What do you think?
Antonin
Hi Antonin,
the -noasegfusion is not really supported and a left over from initial testing when we first developed the stream. I will take a look at it and probably remove the flag (or implement it in a way that it really initializes with the aseg from the base, although I think this is not desirable, unless one really assumes there is no/ or very little change between time points. About aseg edits in longitudinal processing, please look at our edit wiki page: https://surfer.nmr.mgh.harvard.edu/fswiki/LongitudinalEdits#aseg.mgz https://surfer.nmr.mgh.harvard.edu/fswiki/LongitudinalEdits#aseg.mgz
“ ASEG edits can be done in cross, base and long.
Recommendation: Edit in long only if you want to correct the volume of important structures. Edit in base if surfaces need to be changed due to incorrect labeling of aseg. Not necessary to edit in cross.
The segmentation in the long runs is initialized with a fused aseg (a weighted average of the cross sectional asegs). Thus any edits done to the cross sectionals are incorporated indirectly into the long stream. But since the fused aseg is used only for the initialization, it can happen that manual edits from the cross get removed again. Therefore, to correct the volume, just edit the long."
So for your ventricle correction, you could either clean up the cross data and hope it caries over to the long (it should), or directly edit the long aseg.
Best, Martin
On 16 Jan 2017, at 23:23, Antonin Skoch a...@ikem.cz wrote:
Dear experts,
I am testing the longitudinal stream with V6beta version.
The recon-all -help says for -noasegfusion:
Do not create 'fused' aseg from the longbase timepoints, which would normally be used to initialize the ca_labeling. Instead, initialize using the longbase aseg.mgz.
As I looked to recon-all code it seems that in case of -noasegfusion there is no use of -r and -ri parameters and therefore no "initialization". Could you please comment on?
My second query is regarding manual edits:
In one subject the manual editing of aseg was necessary due to ventricle mislabeling. I edited the aseg.presurf in base template and subsequently run longitudinal stream via
recon-all -long tpID templateID -noasegfusion -all
From the last sentence of -noasegfusion help I (wrongly) expected that my aseg.presurf edits in base template would be incorporated for -long recon. However, this was not the case, the edits have not been incorporated at all. Looking at the recon-all code the aseg.presurf from base is never used for -long.
I thought that I could incorporate manual edits from base by manual copying of aseg.presurf from base to long directory. As I looked for the way how the manual aseg edits are handled in recon-all, they are identified by using the difference between the files aseg.auto and aseg.manedit which contains manual edits. Therefore I think that the copying from base is not an option, I expect that this would replace segmentations not only in manual edit regions, but in all regions where the aseg.auto of base and aseg.auto of -long differs (they do not have the same input) which is undesirable. So I think I do not have any other option than to edit aseg.presurf in -long again.
Could you please comment on how to optimally handle such situations?
Regards,
Antonin Skoch _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Hi Antonin,
it would help with editing, but the problem is that we decided initialising from a single aseg is not working in situations with longitudinal changes and non-linear deformations because structures change/move too much. So accuracy will suffer and we decided to allow more flexibility (and more accuracy) at the cost of reliability by initialising with a fuzed aseg. The fuzed aseg is special for each time point, it is not a joint fixed image, but usually looks very similar to the cross aseg of each time point with some small fine tuning when enough other time points agree to vote for a switch in labels.
Initializing with a fixed aseg from the base is only meaningful in test-retest scenarios or situations with very little temporal changes.
Best, Martin
On 17 Jan 2017, at 23:14, Antonin Skoch ansk@ikem.cz wrote:
Dear Martin,
one more thing: When the error is large and present in both timepoints (as is my case), I think that necessity of editing the same error in both timepoints can easily lead to error (maybe even demonstrated as false positive effect) caused by non-systematic editing. When editing comprises many voxels one cannot realistically assure the perfect consistence in editing.
Editing only once (in base) and initialize mri_ca_label in long from base can avoid this problem. So I think the initialization of mri_ca_label from base is not a bad option for such cases.
What do you think?
Antonin
Hi Antonin,
the -noasegfusion is not really supported and a left over from initial testing when we first developed the stream. I will take a look at it and probably remove the flag (or implement it in a way that it really initializes with the aseg from the base, although I think this is not desirable, unless one really assumes there is no/ or very little change between time points. About aseg edits in longitudinal processing, please look at our edit wiki page: https://surfer.nmr.mgh.harvard.edu/fswiki/LongitudinalEdits#aseg.mgz https://surfer.nmr.mgh.harvard.edu/fswiki/LongitudinalEdits#aseg.mgz <https://surfer.nmr.mgh.harvard.edu/fswiki/LongitudinalEdits#aseg.mgz https://surfer.nmr.mgh.harvard.edu/fswiki/LongitudinalEdits#aseg.mgz>
“ ASEG edits can be done in cross, base and long.
Recommendation: Edit in long only if you want to correct the volume of important structures. Edit in base if surfaces need to be changed due to incorrect labeling of aseg. Not necessary to edit in cross.
The segmentation in the long runs is initialized with a fused aseg (a weighted average of the cross sectional asegs). Thus any edits done to the cross sectionals are incorporated indirectly into the long stream. But since the fused aseg is used only for the initialization, it can happen that manual edits from the cross get removed again. Therefore, to correct the volume, just edit the long."
So for your ventricle correction, you could either clean up the cross data and hope it caries over to the long (it should), or directly edit the long aseg.
Best, Martin
On 16 Jan 2017, at 23:23, Antonin Skoch <a...@ikem.cz mailto:a...@ikem.cz> wrote:
Dear experts,
I am testing the longitudinal stream with V6beta version.
The recon-all -help says for -noasegfusion:
Do not create 'fused' aseg from the longbase timepoints, which would normally be used to initialize the ca_labeling. Instead, initialize using the longbase aseg.mgz.
As I looked to recon-all code it seems that in case of -noasegfusion there is no use of -r and -ri parameters and therefore no "initialization". Could you please comment on?
My second query is regarding manual edits:
In one subject the manual editing of aseg was necessary due to ventricle mislabeling. I edited the aseg.presurf in base template and subsequently run longitudinal stream via
recon-all -long tpID templateID -noasegfusion -all
From the last sentence of -noasegfusion help I (wrongly) expected that my aseg.presurf edits in base template would be incorporated for -long recon. However, this was not the case, the edits have not been incorporated at all. Looking at the recon-all code the aseg.presurf from base is never used for -long.
I thought that I could incorporate manual edits from base by manual copying of aseg.presurf from base to long directory. As I looked for the way how the manual aseg edits are handled in recon-all, they are identified by using the difference between the files aseg.auto and aseg.manedit which contains manual edits. Therefore I think that the copying from base is not an option, I expect that this would replace segmentations not only in manual edit regions, but in all regions where the aseg.auto of base and aseg.auto of -long differs (they do not have the same input) which is undesirable. So I think I do not have any other option than to edit aseg.presurf in -long again.
Could you please comment on how to optimally handle such situations?
Regards,
Antonin Skoch _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu mailto:Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Antonin Skoch -
Martin Reuter