Preprocess

Import the experimental virtual environment in conda

conda env -f mran.yaml

then enter the environment:

conda activate mran

Process raw pathology images

Sample dataset download: Link 1 Link 2

After downloading, extract example.zip to the MRAN/WSI/ directory, then preprocess the original image:

cd pre
python run_preprocess.py

The storage format of the original data set can refer to the sample data set in MRAN/WSI/example :

MRAN/WSI/dataset_name/*/slide-idx1.svs

...

MRAN/WSI/dataset_name/*/slide-idxn.svs

Because the example dataset comes from TCGA, the first 12 bits of the file name slide_idxi.svs of each image are its case id.

Divide the dataset

cd pre
python pro_csv.py

The format of the label file of the original dataset can refer to MRAN/csv/example/sheet/total.csv:

File Name	Sample Type
slide-idxi.svs	Primary Tumor
slide-idxj.svs	Solid Tissue Normal
....	...

Train and test

train

python main.py

main.yaml is used to set parameters.

test

python test.py

The predict.csv of the directory "output_dir" records the prediction results on the test set.

Interpretability experiment

cd interpretability
python top_bag_top_patch.py

"output_dir"/tb_tp/*/predict.csv records the prediction results on the test set.