3D Augmentation on CBCT image and segmentation maps

1

Hi,

I am trying to implement 3d augmentation on CBCT and the same augmentations on the 3d segmentation map. I am using the torchio library.

Preprocessing - TorchIO

spatial = tio.OneOf({
       tio.RandomElasticDeformation(): 0.4,
       tio.RandomAffine( degrees=10,  translation = 5): 0.6
  },
   p=0.80,
  )

Can someone please help me out to apply the same transformation to the segmentation map too…
Thanks

PS: Hi @fepegar , Thanks for a wonderful library for 3d augmentations and preprocessing. Can you please help me out with this?

2

Hi, @Dhruv_Jain. As this is not a PyTorch question, you can just ask on the TorchIO GitHub Discussions next time: fepegar/torchio · Discussions · GitHub

There are many examples in the documentation showing how to do that. Here’s a snippet:

In [1]: import torchio as tio

In [2]: t1 = tio.ScalarImage('colin27_t1_tal_lin.nii.gz')

In [3]: brain = tio.LabelMap('colin27_t1_tal_lin_mask.nii.gz')

In [4]: subject = tio.Subject(t1=t1, seg=brain)

In [5]: transform = tio.RandomAffine()

In [6]: transformed = transform(subject)

In [7]: subject.plot()

In [8]: transformed.plot()

Figure_1
Figure_11280×960 122 KB

Figure_2
Figure_21280×960 119 KB

As you see, the same transformation is applied to both images. Linear and nearest neighbor interpolation are used by default for the MRI and the label map, respectively.

3

Thanks :slight_smile: I’ll take care for next time…