IS-MIL Training & validation

Prepare

Step 1. prepare for data

Panda data can be download in https://www.kaggle.com/competitions/prostate-cancer-grade-assessment/data

DiagSet data can be download in https://github.com/michalkoziarski/DiagSet

You need to process the WSI into the following format. The processing method can be found in https://github.com/mahmoodlab/CLAM

The CTransPath feature extractor and the pretrained model can be download in https://github.com/Xiyue-Wang/TransPath

In order to train ISMIL, you need to prepare two folders, one with no overlap between patches and the other with an overlap of 0.5

DATA_DIR
├─patch_coord
│      slide_id_1.h5
│      slide_id_2.h5
│      ...
└─patch_feature
        slide_id_1.pt
        slide_id_2.pt
        ...

The h5 file in the patch_coord folder contains the coordinates of each patch of the WSI, which can be read as

coords = h5py.File(coords_path, 'r')['coords'][:]
# coords is a array like:
# [[x1, y1], [x2, y2], ...]

The pt file in the patch_featurefolder contains the features of each patch of the WSI, which can be read as

features = torch.load(features_path, map_location=torch.device('cpu'))
# features is a tensor with dimension N*F, and if features are extracted using CTransPath, F is 768

Step 2. preparing the data set split

You need to divide the dataset into a training set validation set and a test set, and store them in the following format

SPLIT_DIR
    test_set.csv
    train_set.csv
    val_set.csv

And, the format of the csv file is as follows

slide_id	label
slide_id_1	0
slide_id_2	1
...	...

Train model

Step 1. create a config file

We have prepared two config file templates (see ./configs/) for ISMIL and other baselines, like

General:
    seed: 7
    work_dir: WORK_DIR
    fold_num: 4

Data:
    split_dir: SPLIT_DIR
    data_dir_1: DATA_DIR_1 # for data with no overlap
    data_dir_2: DATA_DIR_2 # for data with overlap
    features_size: 768
    n_classes: 2

Model:
    network: 'ISMIL'

Train:
    mode: repeat
    lr: 3.0e-4
    reg: 1.0e-5
    CosineAnnealingLR:
        T_max: 30
        eta_min: 1.0e-7
    Early_stopping:
        patient: 10
        stop_epoch: 30
        type: max
    max_epochs: 100
    train_method: ISMIL
    val_method: ISMIL
    dataset: TwoStreamBagDataset

Test:
    test_set_name:
        data_dir_1: TEST_DATA_DIR_1 # for data with no overlap
        data_dir_2: TEST_DATA_DIR_2 # for data with overlap
        csv_path: TEST_SET_CSV_PATH

In the config, the correspondence between the Model.network, Train.training_method and Train.val_method is as follows

Model.network	Train.training_method	Train.val_method
ISMIL	ISMIL	ISMIL

Step 2. train model

Run the following command

python train.py --config_path [config path] --begin [begin index] --end [end index]

--begin and --end used to control repetitive experiments

When the training is finished, the code creates a directory with the same name as the config file in the working directory to store all the experimental data. Like this

Task_DIR
│  CONFIG_FILE.yaml
│  s_0_checkpoint.pt
│  s_1_checkpoint.pt
│  s_2_checkpoint.pt
│  s_3_checkpoint.pt
│  s_4_checkpoint.pt
│
└─results
    │
    └─test_set
            metrics.csv
            probs.csv

Evaluation

Please add information about the test set in the config, like this

Test:
    TEST_SET_NAME:
        data_dir_1: TEST_DATA_DIR_1 # for data with no overlap
        data_dir_2: TEST_DATA_DIR_2 # for data with overlap
        csv_path: TEST_SET_CSV_PATH

And run the command

python eval.py --config_path [config path] --dataset_name [TEST_SET_NAME]

Test result will save at WORK_DIR/CONFIG_FILE_NAME/results/TEST_SET_NAME