search

mahmoodlab / CLAM

Data-efficient and weakly supervised computational pathology on whole slide images - Nature Biomedical Engineering
http://clam.mahmoodlab.org
GNU General Public License v3.0
1.07k stars 350 forks source link

Some artifacts are identified as tissue regions #223

Closed huha149 closed 6 months ago

huha149 commented 9 months ago

When I segment the TCGA data, there will be some artifacts that are identified as tissue regions, and thus divided into patches. How can I fix it. 1 HE

HE

fedshyvana commented 6 months ago

Hi, unfortunately this is a limitation of the current segmentation pipeline. You could try maybe adjusting the area threashold go be higher (which will get rid of small artefacts but also potentially smaller pieces of tissue). Although I think in practice, MIL algorithms are pretty robust to artifacts by learning to attend to informative regions (as long as artifacts are not correlated with disease labels).

huha149 commented 6 months ago

Thanks for replying and have a good life