Sorry for not responding earlier; I was away last week.
The fact that MT-CSD comes out OK suggests that your data and response functions are fine and
tckglobal should be able to get a good result too. Given that it fails so dramatically points to the parameter settings of the Metropolis-Hastings optimizer (assuming you are calling the command correctly). Based on what I can see in your screenshots, here are a few steps to try:
-length to 0.5 seems very appropriate for your data.
- I would keep
-weight at the default 0.1 for now. (By setting it to 0.01 the no. degrees of freedom increases by a factor 10, making the optimization more complicated.)
- Do make sure that you run
tckglobal with enough iterations. I typically recommend
-niter 1e9 for adult full brain data, but you may need more depending on the voxel size and segment weight.
- Then try gradually increasing the particle potential
-ppot in a range 0.05 – 0.5.
- If this doesn’t help, you may need to tune the start and end “temperature”
If you want, send me one of your datasets and I’ll have a look.