search

HKU-MedAI / WSI-HGNN

[CVPR'23] Histopathology Whole Slide Image Analysis with Heterogeneous Graph Representation Learning
66 stars 6 forks source link

About the ./checkpoints and ./data file folders #3

Closed WkEEn closed 9 months ago

WkEEn commented 1 year ago

Hi! I will be very appreciated if you can upload the ./checkpoints and ./data to share the weights of pretrianed hovernet and Kiamianet with their json files.

trinhvg commented 10 months ago

Please share your ./data folder, or your ground-truth files, since number of cases in your paper doesn't match with tcga portal, I guess you removed low quality sample?

howardchanth commented 9 months ago

For the weights of Kimianet and HoverNet they can be easily obtained from their original works. Please see #6 for the discussion.

I have uploaded the example label files in the ./data folder. The ground truth is referred to the nationwidechildrens.org_clinicalpatient[DATASET].txt under the clinical data in the TCGA portal. Detail discussions can be found in #5 and #2. For WSI preprocessing, since some (a small number) WSIs are very small and we cannot extract meaningful patches at a certain level, these are removed by the standard tiler we used (i.e., from DSMIL), which is common in WSI preprocessing.