How to prepare unit test images for the pathology tumor detection bundle

[yiheng-wang-nv]

Preparing images for a bundle's unit test is necessary, currently we produced fake input within the test file like:

model-zoo/ci/unit_tests/test_brats_mri_axial_slices_generative_diffusion.py

Line 60 in c2345c1

for s in range(dataset_size):

However, compared with normal images, whole slide images may be more complicated. Hi @drbeh , could you please help to give some suggestions? How to produce fake WSI data for the test? Is there any reference? Thanks in advance!

[Nic-Ma]

Hi @yiheng-wang-nv ,

Please refer to how we prepare the test data in MONAI core:
https://github.com/Project-MONAI/MONAI/blob/dev/tests/test_wsireader.py#L39

Thanks.

[yiheng-wang-nv]

Thanks @Nic-Ma , I will make use of the same tiff data for test.

[drbeh]

Hi @yiheng-wang-nv, we have multiple tiff files for testing in monai core. So far we are using a generic TIFF file and an Aperio SVS file. Both of them are whole slide images and have pyramidical levels and are actually tiff files but the main difference is metadata. For testing purposes, if you solely need the image data, you can use either of them.

[yiheng-wang-nv]

Hi @drbeh , another place to be confirmed is:

model-zoo/models/pathology_tumor_detection/configs/metadata.json

Line 73 in b4d552e

"0": "image"

"channel_def" should describe what are 3 channels stand for. For example:

model-zoo/models/pathology_nuclick_annotation/configs/metadata.json

Line 64 in b4d552e

"channel_def": {

Could you please help to clarify it? Then I can modify this place in the PR:
#455

[drbeh]

Hi @yiheng-wang-nv, for tumor detection bundles channels, I confirm that the channel_def is as follow:

"0": "R",
"1": "G",
"2": "B"