Skip to content

Latest commit

 

History

History
75 lines (64 loc) · 2.8 KB

README.md

File metadata and controls

75 lines (64 loc) · 2.8 KB
Raw

Feature extraction

Here we provide detailed instructions on how to use our mSTAR to conduct feature extraction on WSIs.

Before we start, make sure you have installed the required dependencies mentioned here

Some of the most important libraries are:

timm==0.9.8
openslide==3.4.1
openslide-python==1.3.1

Pipeline

  1. Prepare your WSIs either from your private sources or public sources like TCGA. Download the weights from our Huggingface model page and we recommand to put the weights file in models/ckpts

  2. Follow the instruction from CLAM to preprocess the whole slide images which includes tissue segmentation and stitching. After preprocessing you can get a coordinator folder like this

DIR_TO_COORDS
    └──patches/
        ├── slide_1.h5
        ├── slide_2.h5
        ├── slide_3.h5
        └── ...
  1. Run the following command to conduct feature extraction on WSIs
cd Feature_extract
models='mSTAR'
declare -A gpus
gpus['mSTAR']=0

CSV_FILE_NAME="./dataset_csv/example.csv"

DIR_TO_COORDS="./example_coords"
DATA_DIRECTORY="./example_slide"

FEATURES_DIRECTORY="./output/example"

ext=".jpg"
for model in $models
do
        echo $model", GPU is:"${gpus[$model]}
        export CUDA_VISIBLE_DEVICES=${gpus[$model]}

        python extract_feature.py \
                --data_h5_dir $DIR_TO_COORDS \
                --data_slide_dir $DATA_DIRECTORY \
                --csv_path $CSV_FILE_NAME \
                --feat_dir $FEATURES_DIRECTORY \
                --batch_size 256 \
                --model $model \
                --slide_ext $ext
done

This command will conduct feature extraction on the example WSI we provide. For your own dataset, you need to specify the values for CSV_FILE_NAME, DIR_TO_COORDS and FEATURES_DIRECTORY.

Here is a detailed description of the command line arguments.

  • DATA_DIRECTORY: This should be set to the directory which contains the WSI data.
  • DIR_TO_COORDS: This should be set to the directory that contains the coordinate information for the WSI patches preprocessed through CLAM.
  • FEATURES_DIRECTORY: This is the directory where you want to store the extracted features.
  • model: The feature extractor you want to use, here we use our mSTAR.
  • slide_ext: The WSI file format. For TCGA data, it is .svs, and for the provided example, it is .jpg.

If you run the command above, you will get a folder like this:

output
  └─example
    └─pt_files
      └─mSTAR
        ├── example_1.pt
        ├── example_2.pt
        └── example_3.pt