Crop Labeled ImageData

Use crop() to crop labeled data such as segmented medical images.

from __future__ import annotations

import pyvista as pv
from pyvista import examples

Load a dataset with a CT image and corresponding segmentation labels. Here we load download_whole_body_ct_male().

dataset = examples.download_whole_body_ct_male()

Get the ImageData for the CT data and one of the segmentation masks. For this example we choose a mask of the skull.

ct = dataset['ct']
skull = dataset['segmentations']['skull']

Crop the CT image using the segmentation mask. Use padding to include additional data points around the masked region.

cropped_ct = ct.crop(mask=skull, padding=10)

Use points_to_cells() to plot the cropped image as VOXEL cells.

cpos = [(687.5, 763.6, 471.3), (231.8, 296.3, 677.0), (0.107, 0.311, 0.944)]

cropped_ct_voxels = cropped_ct.points_to_cells()
cropped_ct_voxels.plot(volume=True, cpos=cpos)

Static Scene

Interactive Scene

Include a surface contour of the mask with the plot.

skull_surface = skull.contour_labels()

pl = pv.Plotter()
pl.add_mesh(skull_surface, color='white')
pl.add_volume(cropped_ct_voxels)
pl.camera_position = cpos
pl.show()

Static Scene

Interactive Scene

After cropping, the CT image’s dimensions are smaller than the mask’s.

cropped_ct.dimensions == skull.dimensions

False

To keep dimension the same, either

1. crop the mask itself; the meshes will have smaller dimensions relative to the input

2. pad the CT image as part of the initial crop; the meshes will have the same dimensions as the input

To crop the mask itself, you can perform a similar crop as before using mask=True.

cropped_skull = skull.crop(mask=True, padding=10)
cropped_skull.dimensions

(71, 84, 100)

However, computationally it is more efficient to crop using extent directly.

cropped_skull = skull.crop(extent=cropped_ct.extent)
cropped_skull.dimensions

(71, 84, 100)

Alternatively, use keep_dimensions and fill_value when initially cropping the image so that the output dimensions match the input. A value of -1000 is used, which may represent air in the scan.

cropped_ct = ct.crop(mask=skull, keep_dimensions=True, fill_value=-1000)
cropped_ct.dimensions

(160, 160, 273)