Data_Driven_Color_Augmentation

Implementation of "Data-driven color augmentation for H&E stained images in computational pathology".

Reference

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

[1] Marini N., Otálora S., Wodzinski M., Tomassini S. Dragoni A.F., Marchand-Maillet S., Dominguez P., Duran-Lopez L., Vatrano S., Müller H. & Atzori M., Data-driven color augmentation for H&E stained images in computational pathology.

Paper link: https://www.sciencedirect.com/science/article/pii/S2153353922007830

Requirements

Python==3.6.9, albumentations==0.1.8, numpy==1.17.3, opencv==4.2.0, pandas==0.25.2, pillow==6.1.0, torchvision==0.8.1, pytorch==1.7.0

CSV Input Files:

CSV files are used as input for the scripts. For each partition (train, validation, test), the csv file has path_to_image, class_label as columns. For prostate experiments, the class_label can be: 0: benign 1: Gleason pattern 3 2: Gleason pattern 4 3: Gleason pattern 5

For colon experiments, the class_label can be: 0: cancer 1: dysplasia 2: normal glands

Augmentation

Methods to perform data drive color augmentation (Augmentation.py):

• new_color_augmentation (HSC color augmentation):
  • patch_np: numpy array for the input patch (224x224)
  • kdtree: the database where acceptable color variations are stored
  • alpha: neighbors
  • beta: radius
  • shift_value: perturbation to apply to Hue, Saturation, Contrast
• new_stain_augmentation (perturbation of H&E channels):
  • patch_np: numpy array for the input patch (224x224)
  • kdtree: the database where acceptable color variations are stored
  • alpha: neighbors
  • beta: radius
  • sigma1: range (-sigma1, sigma1) to generate random value to multiply to H&E components
  • sigma2: range (-sigma2, sigma2) to generate random value to add to H&E components

Database

Database including color variations: https://zenodo.org/record/7505727#.Y7ayO3bMJPY.

Method to extend database with new histopathology patches (extend_kd_tree_offline)

• -i: input pickle file (database to extend)
• -o: output pickle file
• -d: csv including patches to extend database

Training

Scripts to train the CNN at path-level, in a fully-supervised fashion. Some parameters must be manually changed, such as the number of classes (output of the network).

• Training_new_augmentation.py -n -b -c -e -f -i -o -a -d. The script is used to train the CNN without any augmentation (no_augment), with colour augmentation (augment).
  • -n: number of the experiment for the training
  • -b: batch size (32)
  • -c: CNN backbone to use (densenet121)
  • -e: number of epochs (10)
  • -t: task of the network (no_augment, augment, normalizer)
  • -f: if True an embedding layer with 128 nodes is inserted before the output layer
  • -i: path of the folder where the input csvs for training (train.csv), validation (valid.csv) and testing (test.csv) are stored
  • -o: path of the folder where to store the CNN’s weights.
  • -a: new augmentation to use: color (HSC color augmentation), stain (H&E stain augmentation), he (H&E-adversarial CNN + HSC color augmentation)
  • -x: extend (False): add color variations training data to color variation dataset
  • -d: database: database including color variations

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.