marugoto_mods

My modifications to marugoto by Kather Lab (https://github.com/KatherLab/marugoto.git)

Also includes a version of highres-WSI-heatmaps (https://github.com/KatherLab/highres-WSI-heatmaps.git) that uses canny edge detection to remove artefacts.

Pre-processing script now includes an artefact detector to remove some tiles that get past the canny edge detector, modified from https://github.com/KatherLab/preprocessing-ng/tree/gecco_cohert.

Please reference their work if using this repo!

NOTE: this is a work in progress - additional functionality may be added as I discover things I need!

Usage

Different scripts in this repo are used in different ways (for legacy reasons).

mil_mods, vis_mods

These two are designed to be imported into another python script. For example:

from marugoto_mods.mil_mods import train_categorical_model_

train_categorical_model_(
        clini_table = clini,
        slide_csv = slide,
        feature_dir = feat_dir,
        target_label = target,
        output_path = outpath,
        lr = lr
    )

Further information can be found in the original repo: https://github.com/KatherLab/marugoto.git

The other files should be run from the command line as follows.

pre-processing

Move to the marugoto_mods folder.
Activate an appropriate venv.
Run the following command:

python -m marugoto_mods.pre-processing \
    --outdir PATH/TO/SAVE/TILES/TO \
    --tile-size INT \
    --um-per-tile FLOAT \
    --brightness-cutoff INT \
    PATH/TO/WSIs

Defaults are:

tile_size = 224
um_per_tile = 256.0
brightness_cutoff = 224

ViT_extractor

Move to the marugoto_mods folder.
Activate an appropriate venv.
Run the following command (augmented_repetitions is optional):

python -m marugoto_mods.ViT_extractor \
      --slide_tile_paths SLIDE_TILE_DIR \
      --outdir PATH/TO/SAVE/FEATURES \
      --augmented_repetitions INT

gradcam

This python file contains the functions required to apply the gradcam technique to 1 the features for a WSI and 2 the top tiles within said tile. The GCAM function saves 3 images (a thumbnail of the slide, slide-level gradcam result and tile-level gradcam result) to the specified outdir.

python -m marugoto_mods.gradcam \
  model = PATH/TO/MARUGOTO_MODEL/EXPORT.PKL \
  slide_path = PATH/TO/SLIDE \
  feat_dir = PATH/TO/FEATURE/DIRECTORY \
  outpath = PATH/TO/OUTPUT/IMGS/TO

heatmaps

Move to the marugoto_mods folder.
Activate an appropriate venv.
Run the following command:

python -m marugoto_mods.create_heatmaps.py [-h] -m MODEL_PATH -o OUTPUT_PATH -t TRUE_CLASS
                                           [--no-pool]
                                           [--mask-threshold THRESH]
                                           [--att-upper-threshold THRESH]
                                           [--att-lower-threshold THRESH]
                                           [--score-threshold THRESH]
                                           [--att-cmap CMAP]
                                           [--score-cmap CMAP]
                                           SLIDE [SLIDE ...]

Further information available at https://github.com/KatherLab/highres-WSI-heatmaps