search

binli123 / dsmil-wsi

DSMIL: Dual-stream multiple instance learning networks for tumor detection in Whole Slide Image
MIT License
378 stars 88 forks source link

TCGA Labels #75

Closed jafarinia closed 1 year ago

jafarinia commented 1 year ago

Hi. Thanks for the wonderful work. Can I know where exactly in TCGA website you find the label for TCGA WSIs being cancer/healthy. Because I can't find this information on their website or any data that you provide .Only your csv files for labels only says if the WSI is LUAD or LUSC and later you just decide based on the attention weights if a patch and finally a slide is cancer or not (which I don't know if it's the right thing to do especially because there is no way to test it)

binli123 commented 1 year ago

https://portal.gdc.cancer.gov/ Lung -> disease type (LUSC or LUAD) -> file type svs. There is no evaluation of the patches for the TCGA dataset due to there being no ground truth patch-level labels. There is also no claim that the attention score for the TCGA can be used to DECIDE whether a slide is cancer or not. The slide classification is done at the aggregated slide level. The attention directs the attended area that may be used for further examination for tumour type.