[feature request] N4 bias correction support

[LennartPhil]

Is your feature request related to a problem? Please describe.
N4 bias correction is quite a common preprocessing step when it comes to MRI image data analysis, which is why it should be part of the BrainLesion preprocessing tool. Adding this step manually at the beginning or at the end does not reflect the idea of a compact preprocessing tool.

Describe the solution you'd like
I'd like to be able to add N4 bias correction to the preprocessing pipeline, this could be achieved for example by using SimpleITK. This could go right with the already added normalization step.

Considering that all the different normalizers are stored in a different Github repository, it might make sense to merge the normalizers into the current repo, to simplify any future changes.

Edit: Z-score normalization should be considered as well. But it should be noted, that z-score normalization especially also requires the brain mask as an input

[neuronflow]

In our experiments, we never found an advantage of running N4 bias field corrections, but I agree, that is well-established in the community. Thus, it should be part of our preprocessing pipeline.

Thank you for the suggestion @LennartPhil

When does N4 produce best results?
should this be run on images with or without skull? before or after normalization?

[neuronflow]

@MarcelRosier, please also see this thread regarding our discussion about normalizers.

[neuronflow]

@LennartPhil I see you created a fork
https://github.com/LennartPhil/preprocessingN4

Would you like to work on a PR?

[sarthakpati]

@neuronflow: I have added this [ref].

[neuronflow]

we believe bias field correction should run before normalization and bet

[sarthakpati]

I have added the functionality, and left it for you to add to the PR based on where you think it should be called.

[MarcelRosier]

@neuronflow: I have added this [ref].

Thanks for the implementation @sarthakpati ! I'm a bit confused tho where the mask is coming from? sarthakpati@b65ba06#diff-11dfdd0e47eb68c97e41d6f1e764a6393836abffd17a772a25e4b6f0197826ceR52

[sarthakpati]

The mask can be provided if one wants to perform the bias correction only in the masked region. It can be computed using whatever logic the user thinks is appropriate (I have seen otsu, percentile, and a hard coded threshold being used for different types of data).