preprocessing with Label about unires

[brudfors]

Hi @jakubMitura14

Nice to hear that you find the code useful :)

Yes, you can include labels (ensuring that the affine matrices are modified correctly):

UniRes/unires/struct.py

Line 85 in 82d00af

self.label: tuple = None # Manual labels, given as (str, (int, int))),

For example, if you have a label file label.nii.gz and one modality you would specify it as:

from unires.struct import settings as s

s.label = ('label.nii.gz', (0, 0))

This function might give some insights into using unires as an api:

UniRes/unires/_cli.py

Line 7 in 82d00af

def _preproc(pth, atlas_rigid, common_output, device, dir_out, fov, label_file,

Keep in mind that your prostate images need to have good alignment between subjects as unires internal inter-subject alignment method currently only works for brain scans (it uses an atlas-based approach with a T1w brain mean image). That is, this option:

UniRes/unires/struct.py

Line 77 in 82d00af

self.do_atlas_align: bool = False # Align images to an atlas space

needs to be set to False. However, if you have some prostate mean image, it could be used to perform the inter-subject alignment.

[jakubMitura14]

Fantastic! thanks for your response now I will work on it !

[jakubMitura14]

I would like to ask something futher - precisely about "common_output" - it seem fantasic to get all of the data on the same grid - hovewer I do not see exactly how it works?
I seen that I can pass into the preproc function two dimensional list of paths to nifti files - but only one setting file
so I should put the labels into tuple like

s.label = ('labelA.nii.gz','labelB.nii.gz' )... ?

Also I have t2 (axial coronal, saggital) and adc images labels are associated with t2w image (it was annotated basically on t2 axials) - how to mark it so during registration and resampling labels will be adjusted as the t2w?

Lastly I had done experiment and all resampled files have been written - but not labels, how to specify where the modified resampled labels should be present?

Thanks for Help, and by the way tool is fantastic - It solves so many preprocessing problems at once that it seems automagical !

[brudfors]

If you have three images of the same 'repeat' (three T2 scans of different thick-slice direction: coronal, sagittal, axial) and the label is annotated on the axial scan, you would specify this as:

data = [['t2_c.nii.gz', 't2_s.nii.gz', 't2_a.nii.gz']]
s.label = ('label.nii.gz', (0, 2))

You have to ensure that your label image is an integer tensor with values {0, 1, ..., num_labels}. Meaning that if you have multiple binary masks, you will need to merge these into a categorical one.

[jakubMitura14]

perfect, Thanks! and if I want to put whole dataset (given I had done earlier interpatient registration ) on the same grid as described in "common_output" how to pass multiple labels? And is it problem that in some cases label will be present and in other not - only part of dataset is annotated

Thank You!

[brudfors]

UniRes runs on one subject at a time, not on a population. So you would have to call it multiple times, once for each subject.

[brudfors]

If no labels are available, then you would just call UniRes w/o labels for those subjects.