How to get segmentations as numpy arrays with same dimensions?

Support
1

hello, lassoan. I encountered the same problem when save the segmentation result to file. I right clicked the segmentation in the “Subject hierarchy”, and then choose “export visible segments to binary labelmap”, and got the labelmap which name is “Segmentation-bone-label”. And then I got the segmentation mask array using python. But the size of each image in segmentation mask is only 224332, and the original image in volume is 512512. I got the segmentation mask which size is same as original image before. And I donot know why this time the size is only bounding box. Here is the python code I used:

bone=array(“Segmentation-bone-label”)
bone.shape
(114, 224, 322)
volume = array(“5 ARTERIAL PHASE”)
volume.shape
(114, 512, 512)

Looking forward to your reply, thank you so much.

2

The simplest is to use the latest Slicer version, which does not crop the segmentation to the minimum necessary size. If you already created your segmentation in Slicer-4.10 then you need to specify a Reference volume as described above.

4

The version is 4.11(2020-6-15). And where can i find the advanced export section in 4.11 to add reference volume. In the screenshot below, you can see the segmentation size is not 512*512, and the Slicer version is 4.11.

export
export816×1055 49.1 KB

5

How did you create Segmentation_1-bone-label and Segmentation_2-bone-label?
If they were created from two different segmentations (of two different volumes), could you check the dimensions of these volumes (in Volumes module, Volume Information section)?

6

In addition to @lassoan replies, you can change the segmentation geometry to match volume geometry with SetReferenceImageGeometryParameterFromVolumeNode(). This allowed me in the past to batch process mismatched examples, e.g.

loadedVolumeNode = slicer.util.loadVolume(ct_path)
loadedSegmentationNode = slicer.util.loadSegmentation(seg_path)
loadedSegmentationNode.SetReferenceImageGeometryParameterFromVolumeNode(loadedVolumeNode)
slicer.util.saveNode(loadedSegmentationNode, seg_path)
slicer.mrmlScene.RemoveNode(loadedVolumeNode)
slicer.mrmlScene.RemoveNode(loadedSegmentationNode)
3 Likes
7

yes, they are from differnent volumes.Here is how can I create the labelmap in the scrrenshot 1-2. and the third screenshot showes the volume dimension. And I tried to get the dimension of the labelmap, the result is shown below.

bone = array(“Segmentation_5-bone-label”)
bone.shape
(118, 276, 382)
volume = array(“4 Tx Abdomen 5mm”)
volume.shape
(118, 512, 512)

binary
binary1920×1023 272 KB

8

I found if I save the segmentation to nrrd file first, and then load it and volume again. and then the labelmap created from the segmentation is bounding box size. If I save the segmentation directly after segmentation, the labelmap size is same as the volume(512*512).

9

Your solution works greatly. Thanks a lot. :+1:
BTW, do you know how to get the node name (like loadSegmentationNode and loadedVolumeNode in your code) if I load the volume and segmentation from GUI. thanks again.