Resample in one dimension only?

[scottassen]

Is there a way to resample an image in one pixdim/dimension only?

For example, in the spleen segmentation dataset, the original image shapes range from (512, 512, 31) to (512, 512, 168), with axial (z) voxel dimensions of 1.5 mm to 8 mm. Is there a way to resample these images to all have the same axial (z) voxel dimensions, while leaving the x and y dimensions the same (so that the final image shapes are still all (512, 512, z))?

The Spacing transform doesn't let you keep original pixdims for some dimensions, and Resize would make all images exactly the same dimensions.

Example desired input:

1. image shape: torch.Size([1, 512, 512, 60])
   pixdim: [1.       0.751953 0.751953 5.       0.       0.       0.       0.      ]
2. image shape: torch.Size([1, 512, 512, 31])
   pixdim: [1.       0.785156 0.785156 7.5      0.       0.       0.       0.      ]

Example desired output:

1. image shape: torch.Size([1, 512, 512, 60])
   pixdim: [1.       0.751953 0.751953 5.       0.       0.       0.       0.      ]
2. image shape: torch.Size([1, 512, 512, 47])
   pixdim: [1.       0.785156 0.785156 5      0.       0.       0.       0.      ]

[Nic-Ma]

Hi @scottassen ,

Thanks for your feature request, yes, we may consider to support pixdim=(-1, -1, X) for your case.
In your particular example, I think you can use Spacingd + Resized transforms.
Just set spatial_size=[512, 512, -1] in Resized to keep the last dim after Spacingd.
https://github.com/Project-MONAI/MONAI/blob/dev/monai/transforms/spatial/array.py#L330

Thanks.

[Nic-Ma]

I submitted PR to support this feature request: #2423.

Thanks.

[scottassen]

Thank you! And I appreciate the tip to use Spacingd and Resized in order.