This page contains a list of common issues that might arise during MICe-build-model usage and how to solve them. It is currently very much under construction, but will be updated in the coming weeks and months.
Q: Why do my in-vivo brains not align properly using large rotations in lsq6?
A: There can be several answers to this problem. But most likely your input file, and the target (native.mnc from your initial model) do not look similar enough. Two examples would be:
- Your file is "wrapped around" (see below). Solution: reconstruct your image such that the entire brain and tissue around the brain are attached (as is the case in the target)
The position of the brain in your input file is in a (quite) different location as your target (see below). Solution: use a mincreshape command to artificially move the brain in a more similar location. In the example below, the following was done:
Q: When running the lsq12 portion of the registration, some or all pairwise registrations fail. When you check the log files, the error message contains the following line:
No voxels were found in volume 2 with value above threshold (0.000000).
Why is this happening?
A: minctracc, the registration software that is used for the lsq12 (12 parameter, affine) registration part of the alignment process assumes that the volumes it compares contain voxels that have values above 0. If it does not find any voxels with a positive intensity it will exit with this error message. The way you can visually check what is wrong is by looking at the input files for the minctracc command. In the same log file where you found the error, the minctracc command that was used is stated and will have the form:
minctracc -clobber -lsq12 -w_translations 0.4 0.4 0.4 -simplex 3 -verbose 2 -step 1 1 1 -tol 0.0001 -xcorr -identity -source_mask mask.mnc -model_mask mask.mnc source.mnc target.mnc transform.xfm
You can pull up both that source.mnc and target.mnc file in one of the visualization programs (OCCIviewer, register, Display) to see what the problem is. Two possible options:
- If you used -inormalize and are running a registration pipeline with masked input files, it is possible for the normalization step to go wrong such that the contrast in the images is flipped. Visually you will see that the background looks white instead of black. If you check the intensities in the brain you will see that they are negative. To fix this, rerun your registration with the -no-inormalize flag.
- It is possible for the lsq6 stage to go wrong, and a brain could have been "rotated out of the field of view". You might see a bit of the brain in the corner of the volume, but most of it will be empty (black). In that case you can manually register the brain using 6 parameters (in register using 4 or 5 landmarks) to you native.mnc file from your initial model folder.