[FEATURE] Adjustment for CT image defacing

[nicmuenster]

Currently the defacing works by multiplying a mask consisting of 1s and 0s to the image. However, in the case of CT images, the intensity range can have a minimum that is far below 0, which can lead to this newly masked region being portrayed very differently to the actual background. Therefore it might worth considering to allow to define a value that is inserted into this region.

This could for example be done by changing the line

masked_data = input_data * mask_data

to

masked_data = input_data * mask_data + ((1 - mask_data) * self.fill_value)

where fill_value is decided upon during the initialization of the defacer.

[neuronflow]

Nice catch @nicmuenster. Would you like to submit a PR for this?

Should we do the same procedure for skullstripping masks?

PS: I have a feeling that there might be a more elegant and readable way to implent this via subsetting.

[neuronflow]

something like:

masked_data = np.where(mask_data, input_data, self.fill_value)

or:

masked_data = input_data.copy()
masked_data[mask_data == 0] = self.fill_value