Skip to content

How to use smaller patch size for manual segmentation #312

Open
@NiekPennings

Description

@NiekPennings

Hey all!

I have a question about the usage of CLAM.
I have a dataset of about 500 histological images of brain regions with iron staining. These images are already downscaled by a factor of 20x and then segmented. This means that I am not using pyramidal TIFFS but already segmented 3000x3000 jpgs. I have noticed that this framework is not entirely designed for this use case, but I think it will work fine!

I'm currently struggling with tweaking the parameters for this specific dataset. Using patches of 256x256 and a line thickness of 250 just creates ridiculous images like this one below

Image

That is why I want to use patches of about 16x16. However, I don't seem to be able to make that work with the create_patches_fp.py. Currently, it seems to make a grid of 16 patches of where the original tissue seems to be, like this.

Image

The parameters I have used for this result is the bwh_resection preset, with a patch size of 16.

In general, I feel a bit lost when it comes to changing these parameters in the script.

  • The seg_ and vis_level are not so interesting for me, as I only have 1 image.
  • The line thickness should not be much more than 5 pixels, I think
  • max_n_holes is not that interesting for me, as I don't have that many holes in my segmentation. It is mostly one piece of tissue.

For the rest of the parameters, I feel a bit clueless as to what it exactly does. Although it feels like the parameters don't really change that much, and I just end up with the small pixel grid all the time.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions