Digital_Pathology_Transfer_Learning

Implementation of "Combining weakly and strongly supervised learning improves strong supervision in Gleason pattern classification"

If you find this code useful, consider citing the accompanying article: Sebastian Otálora, Niccolò Marini, Manfredo Atzori, and Henning Müller, et al. "Combining weakly and strongly supervised learning improves strong supervision in Gleason pattern classification"

Requirements

Python==3.6.9, tensorflow==1.14.0, numpy==1.17.3, opencv==4.2.0, pandas==0.25.2, pillow==6.1.0, keras==2.2.4, staintools==2.1.2

Datasets

Two datasets are used for the experiments:

• The Tissue Micro Array Zurich (TMAZ)
• The Cancer Genome Atlas-PRostate ADenocarcinoma (TCGA-PRAD)

The lists of images used (split in training, validation, testing partitions) are in:

• TMAZ_images
• TCGA_images

Folder organization

csv_files

It inclused the csv files. The csv files MUST BE in this format: filename, label.

test

It includes the scripts to test fully-supervised, weakly-supervised, finetuning models at patch level (TMAZ), core level (TMAZ) and at whole slide image level (TCGA-PRAD).

train

It includes the scripts to train fully-supervised, weakly-supervised, finetuning models.

utils

It includes scripts used to extract patches from TMAs/WSIs and scrits to generate the csvs.

Scripts organization

The repository includes the scripts for training the models (training_scripts), for testing the models (inference scripts) and for the preprocessing/generation of the csv files (utils).

train:

• Fully_Supervised_training.py -n -p. The script is used to train the model with fully-supervision.

  • -n: number of experiment
  • -p: percentage of data to use (20,40,60,80,100).

• Weakly_Supervised_training.py -n The script is used to train the model with weakly-supervision.

  • -n: number of experiment.

• Finetuning_with_weakly_annotated_data.py -n -p. The script is used to finetune the model (pre-trained with fully-supervision) with weakly-annotated data.

  • -n: number of experiment
  • -p: percentage of data used to pre-train (20,40,60,80,100).

test:

• Inference_TMA_patches.py -n -e -p. The script is used to test the models at patch-level (TMAZ).

  • -e: method of train (options: fully/weak/finetune).
  • -n: number of experiment to test.
  • -p: percentage of data used to pre-train (20,40,60,80,100).

• Inference_TMA_cores.py -n -e -p. The script is used to test the models at core-level (TMAZ).

  • -e: method of train (options: fully/weak/finetune).
  • -n: number of experiment to test.
  • -p: percentage of data used to pre-train (20,40,60,80,100).

• Inference_TCGA_densely_keras.py -n -e -p. The script is used to test the models at WSI-level (TCGA-PRAD).

  • -e: method of train (options: fully/weak/finetune).
  • -n: number of experiment to test.
  • -p: percentage of data used to pre-train (20,40,60,80,100).

utils

• TMA_Patch_Extractor.py -d -s -n -t -p. Script to extract the patches from the TMAZ dataset (using pixel-wise annotated masks).

  • -d: dataset where extract patches (train/valid/test).
  • -s: size of the tiles to extract (before the resize to 224x224).
  • -n: number of patches to extract.
  • -t: number of threads.
  • -p: minimum percentage of tissue in a tile to be accepted.

• WSI_Patch_Extractor.py -d -s -t -p. Script to extract the patches from the TCGA-PRAD dataset (using masks generated by HistoQC).

  • -d: dataset where extract patches (train/valid/test).
  • -s: size of the tiles to extract (before the resize to 224x224).
  • -t: number of threads.
  • -p: minimum percentage of tissue in a tile to be accepted.

• Create_percentage_datasets.py -n. Script to create the subset with percentage of data.

  • -n: number of experimet.

• Create_csv_weakly_annotated_data.py -d. Script to create the csv file with weakly-annotated data.

  • -d: partition (train/valid/test)

Reference

If you find this repository useful in your research, please cite:

[1] Otálora, S., Marini, N., Müller, H., & Atzori, M. (2021). Combining weakly and strongly supervised learning improves strong supervision in Gleason pattern classification. BMC Medical Imaging, 21(1), 1-14.

Paper link: https://bmcmedimaging.biomedcentral.com/articles/10.1186/s12880-021-00609-0

Acknoledgements

This project has received funding from the EuropeanUnion’s Horizon 2020 research and innovation programme under grant agree-ment No. 825292 ExaMode. Infrastructure fromthe SURFsara HPC center was used to train the CNN models in parallel. Otálora thanks Minciencias through the call 756 for PhD studies.