What is the default regularization method for tcksift2?

1

Hello MRtrix experts,

What is the default regularization method for tcksift2?
If I run tcksift2 -reg_tv 0.1 ..., would I select the TV regularization?

Thanks,
xiaoping

2

I think this is all stated in the help page for tcksift2:

  • -reg_tikhonov value provide coefficient for regularising streamline weighting coefficients (Tikhonov regularisation) (default: 0)
  • -reg_tv value provide coefficient for regularising variance of streamline weighting coefficient to fixels along its length (Total Variation regularisation) (default: 0.1)

This basically says the default is TV with value 0.1, Tikhonov is not used (value 0).

3

If I run tcksift2 -reg_tv 0.1 ... , would I select the TV regularization?

These options only influence the strength of regularisation for that particular metric. So e.g. if you were to specify -reg_tik #, this would not disable the TV regularisation.

The default of 0.1 on the TV regularisation is perhaps slightly over-regularised, in order to reduce the prevalence of very large fluctuations in streamline weights that can occur due to esoteric details of the optimisation algorithm. You can definitely get a better fit to the image data with e.g. -reg_tv 0.05. But if fiddling with this it’s important to understand both how the mechanism is working, and the ramifications of altering the values used.