Hello Freesurfer Experts-
We recently reran several subjects from scratch through the stable Freesurfer 5.0 pipeline who had previously been processed with the stable 4.5 pipeline. We used tkregister2 for QA of the 5.0 data and noticed that the talairach transform was off (misaligned) for the first 5 subjects we examined. All subjects passed the auto Talairach check, and their 4.5 data was not misaligned.
Applying the talairach.xfm matrix from the 4.5 pipeline corrected the problem. Looking at the recon-all.cmd files from the different versions, it looks like the only difference for the tailairach_avi step was that the 4.5 version called for nu.mgz whereas the 5.0 version called for orig.mgz. I reran a couple of 5.0 subjects using
talairach_avi --i nu.mgz –xfm transforms/talairach.auto.xfm
Doing this fixed the transform and subsequent alignment problems (copied above to talairach.xfm and re-ran tkregister2).
I have read on the list that the talairach.xfm is not used in the majority of data streams (although table indicates otherwise). Our plans were to use the data for both cortical thickness analyses as well as MEG analyses (source mapping). I would appreciate it if someone could tell me:
1) The best way to rerun the current data that has been run through 5.0. We have already run about 50 of 100 subjects through 5.0.
2) The best way to fix our recon-all program for 5.0 so that the correct transform is generated for new subjects.
3) In addition, we also observed the following error in the log file and could not find any information on what it may mean (all checked subjects had error)
***********************************************
Computing MAP estimate using 3481 samples...
***********************************************
IFLAG= -1 LINE SEARCH FAILED. SEE DOCUMENTATION OF ROUTINE MCSRCH
ERROR RETURN OF LINE SEARCH: INFO= 3 POSSIBLE CAUSES: FUNCTION OR
GRADIENT ARE INCORRECT OR INCORRECT TOLERANCESoutof QuasiNewtonEMA:
**************************************************
EM alignment process ...
Computing final MAP estimate using 382743 samples.
**************************************************
IFLAG= -1 LINE SEARCH FAILED. SEE DOCUMENTATION OF ROUTINE MCSRCH
ERROR RETURN OF LINE SEARCH: INFO= 3 POSSIBLE CAUSES: FUNCTION OR
GRADIENT ARE INCORRECT OR INCORRECT TOLERANCESpass 2 through
quasi-newton minimization...
Thanks for your time!
Andrew
Andy,
I'll answer the last first: that error message (the one in ALL CAPS) can be ignored. It originates in the bowels of some fortran code in the VNL library we use. A real error detected at the level of our code would exit if there was a problem with the minimization.
Per the talairach.xfm, as you have found, in 5.0, it uses orig.mgz as input to the talairach stage, which you and others have found to not produce an optimal transform (at least as viewed in tkregister2 compared to that produced by v4.5). In the next release, a change is being made to recon-all to create a temporary nu_correct'd nu for input to the talairach. however, as long as the talairach is decent, then surface creation should be fine. as we recommend, inspect the surfaces for defects. a really poor talairach might result in a bad skull strip, which would result in bad surfaces, but this would be apparent during inspection.
the other need for an 'optimal' talairach.xfm is if you make use of the ICV calculation, which is based on that transform. if you do, then probably you should inspect each talairach.xfm (using tkregister2) and correct problems, and/or rerun the creation of it using nu.mgz as input as you have done. you would not need to rerun the surface stream using this new talairach.xfm.
n.
On Thu, 2011-12-22 at 11:54 -0700, Andy Mayer wrote:
Hello Freesurfer Experts-
We recently reran several subjects from scratch through thestable Freesurfer 5.0 pipeline who had previously been processed with the stable 4.5 pipeline. We used tkregister2 for QA of the 5.0 data and noticed that the talairach transform was off (misaligned) for the first 5 subjects we examined. All subjects passed the auto Talairach check, and their 4.5 data was not misaligned.
Applying the talairach.xfm matrix from the 4.5 pipelinecorrected the problem. Looking at the recon-all.cmd files from the different versions, it looks like the only difference for the tailairach_avi step was that the 4.5 version called for nu.mgz whereas the 5.0 version called for orig.mgz. I reran a couple of 5.0 subjects using
talairach_avi --i nu.mgz –xfm transforms/talairach.auto.xfm
Doing this fixed the transform and subsequent alignmentproblems (copied above to talairach.xfm and re-ran tkregister2).
I have read on the list that the talairach.xfm is not used inthe majority of data streams (although table indicates otherwise). Our plans were to use the data for both cortical thickness analyses as well as MEG analyses (source mapping). I would appreciate it if someone could tell me:
The best way to rerun the current data that has been runthrough 5.0. We have already run about 50 of 100 subjects through 5.0.
The best way to fix our recon-all program for 5.0 so that thecorrect transform is generated for new subjects.
In addition, we also observed the following error in the logfile and could not find any information on what it may mean (all checked subjects had error)
Computing MAP estimate using 3481 samples...
IFLAG= -1 LINE SEARCH FAILED. SEE DOCUMENTATION OF ROUTINE MCSRCH
ERROR RETURN OF LINE SEARCH: INFO= 3 POSSIBLE CAUSES: FUNCTION OR
GRADIENT ARE INCORRECT OR INCORRECT TOLERANCESoutof QuasiNewtonEMA:
EM alignment process ...
Computing final MAP estimate using 382743 samples.
IFLAG= -1 LINE SEARCH FAILED. SEE DOCUMENTATION OF ROUTINE MCSRCH
ERROR RETURN OF LINE SEARCH: INFO= 3 POSSIBLE CAUSES: FUNCTION OR
GRADIENT ARE INCORRECT OR INCORRECT TOLERANCESpass 2 through
quasi-newton minimization...
Thanks for your time!
Andrew
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
addendum: i mistakenly said that talairach.xfm is used in the skull-strip stage... it is not. so beyond really bad talairach failures, the only critical reason to redo the talairach.xfm is if the ICV calculation (which itself is pretty rough) is valuable to you (for correcting volume stats by head size).
n.
On Sat, 2011-12-24 at 20:20 -0500, Nick Schmansky wrote:
Andy,
I'll answer the last first: that error message (the one in ALL CAPS) can be ignored. It originates in the bowels of some fortran code in the VNL library we use. A real error detected at the level of our code would exit if there was a problem with the minimization.
Per the talairach.xfm, as you have found, in 5.0, it uses orig.mgz as input to the talairach stage, which you and others have found to not produce an optimal transform (at least as viewed in tkregister2 compared to that produced by v4.5). In the next release, a change is being made to recon-all to create a temporary nu_correct'd nu for input to the talairach. however, as long as the talairach is decent, then surface creation should be fine. as we recommend, inspect the surfaces for defects. a really poor talairach might result in a bad skull strip, which would result in bad surfaces, but this would be apparent during inspection.
the other need for an 'optimal' talairach.xfm is if you make use of the ICV calculation, which is based on that transform. if you do, then probably you should inspect each talairach.xfm (using tkregister2) and correct problems, and/or rerun the creation of it using nu.mgz as input as you have done. you would not need to rerun the surface stream using this new talairach.xfm.
n.
On Thu, 2011-12-22 at 11:54 -0700, Andy Mayer wrote:
Hello Freesurfer Experts-
We recently reran several subjects from scratch through thestable Freesurfer 5.0 pipeline who had previously been processed with the stable 4.5 pipeline. We used tkregister2 for QA of the 5.0 data and noticed that the talairach transform was off (misaligned) for the first 5 subjects we examined. All subjects passed the auto Talairach check, and their 4.5 data was not misaligned.
Applying the talairach.xfm matrix from the 4.5 pipelinecorrected the problem. Looking at the recon-all.cmd files from the different versions, it looks like the only difference for the tailairach_avi step was that the 4.5 version called for nu.mgz whereas the 5.0 version called for orig.mgz. I reran a couple of 5.0 subjects using
talairach_avi --i nu.mgz –xfm transforms/talairach.auto.xfm
Doing this fixed the transform and subsequent alignmentproblems (copied above to talairach.xfm and re-ran tkregister2).
I have read on the list that the talairach.xfm is not used inthe majority of data streams (although table indicates otherwise). Our plans were to use the data for both cortical thickness analyses as well as MEG analyses (source mapping). I would appreciate it if someone could tell me:
The best way to rerun the current data that has been runthrough 5.0. We have already run about 50 of 100 subjects through 5.0.
The best way to fix our recon-all program for 5.0 so that thecorrect transform is generated for new subjects.
In addition, we also observed the following error in the logfile and could not find any information on what it may mean (all checked subjects had error)
Computing MAP estimate using 3481 samples...
IFLAG= -1 LINE SEARCH FAILED. SEE DOCUMENTATION OF ROUTINE MCSRCH
ERROR RETURN OF LINE SEARCH: INFO= 3 POSSIBLE CAUSES: FUNCTION OR
GRADIENT ARE INCORRECT OR INCORRECT TOLERANCESoutof QuasiNewtonEMA:
EM alignment process ...
Computing final MAP estimate using 382743 samples.
IFLAG= -1 LINE SEARCH FAILED. SEE DOCUMENTATION OF ROUTINE MCSRCH
ERROR RETURN OF LINE SEARCH: INFO= 3 POSSIBLE CAUSES: FUNCTION OR
GRADIENT ARE INCORRECT OR INCORRECT TOLERANCESpass 2 through
quasi-newton minimization...
Thanks for your time!
Andrew
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu 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
-
Andy Mayer -
Nick Schmansky