digpath_msi_prediction

Steps for getting patch embeddings

1. get_filenames.sh

   • Set TOP_LEVEL_DIR to the directory containing the slide-specific subdirectories
   • run with ./get_filenames.sh <label> replacing "label" with 0 or 1 (usually MSS=0, MSI=1)

2. get_features_CTransPath.py

   • Fill in line 51 with your path to ctranspath.pth
   • Run with python get_features_CTransPath.py <filenames.csv> replacing filenames.csv with the output csv of get_filenames.sh
   • May want to write a shell script to run for all slides' csv files

3. make_h5.py

   • Set directory to the path to the directory containing the embedding CSVs
   • Set out_directory to the desired output directory for the h5 files
   • Set output_csv to the desired output file for the slide.csv file (an input to HistoBistro)