Jacobian Limit Fix

[clev...@umn.edu]

Hello,

I am attempting to use DTI-TK to create a study-specific target tensor image, using the IIT atlas as a base. I first want to create a population template of my chosen subjects, then register this to the IIT atlas. While generating the population template, in some subjects, dti_diffeomorphic_reg terminates early, after only four or five iterations, passing the following error:

JACOBIAN STATISTICS: after previous iteration mean = 0.99641 min = 0.0608428 max = 5.01594 # of voxels = 1.98676e+06 after current iteration mean = 0.995373 min = 0.0169021 max = 6.56487 # of voxels = 1.98676e+06
iteration 4 done, not accepted ...
if you believe the registration terminates too early, you can relax the jacobian limit.
default jacobian limits can be changed by modifying dtitk_common.sh in the scripts directory of DTI-TK.

After looking through the list, it seems the most typical reason for this error is that the data lands outside the bounds of the mask. However, the subjects in question appear normal with respect to the others in the set, and do not appear to have a significantly greater offset from the mask. I have attempted to inflate the mask, erode the data, or do both simultaneously. In all cases, dti_diffeomorphic_reg still terminates early and passes the same error for these subjects. When I dropped a problem subject from the group template, a different subject which had previously run correctly exhibited the same error. Again, this subject appeared normal with respect to others in the pool.

What other issues could be leading to the unaccepted iterations? Under what conditions should we consider relaxing the jacobian limits for all subjects? And if this happens when running through a standard pipeline, are these subjects effectively unusable and should be dropped, or can they be salvaged?

Thank you for your help.

Best,

Duncan Cleveland

University of Minnesota

[Hui Zhang]

Hi Duncan,

This problem often occurs near the top and/or bottom part of the volumes.  The deformation fields we estimate need to be zero at the boundary of the volumes.  If part of your mask gets really close to the volume boundary, this can force the deformation fields to change rapidly.  So the solution is to reduce the mask a bit.  Not eroding but just clearing a slice or two near the edge of the volume.

Hope this makes sense.

Gary

--

---
You received this message because you are subscribed to the Google Groups "DTI-TK" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dtitk+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.