Data collection and preprocessing

To ensure reproducibility of the results, coordinates of the tiles used in the paper (necessary to extract tile images and features from whole-slide images) can be downloaded from https://drive.google.com/file/d/1PJsUv1SQieJs7hqtWOqW68v1K9c-mIF6/view?usp=sharing. Activity signature values can be found in assets/activity_signature and the python formula is in he2tead/preprocessing/signature_computation.py

Install

To install openslide, do:

apt-get update -qq && apt-get install openslide-tools libgeos-dev -y 2>&1

Then to install he2tead:

pip install -e .

Dataset

The general class to load data is TCGADataset present in he2tead/data/dataset.py.

Feature Extraction

ResNet50 pre-trained with supervised learning on ImageNet dataset

In the file he2tead/preprocessing/imagenet_extraction.py, we provide an example on how to extract features from each tiles, given their coordinates, using a ResNet50 pre-trained on ImageNet dataset.

Wide ResNet50 x2 pre-training with MoCo-v2 on histology images

The feature extractor used in our study is a Wide ResNet50 x2, that was pre-trained with MoCo v2 on 4 million tiles from TCGA-COAD dataset.

The code to train such model is available here: https://github.com/facebookresearch/moco.

Other feature extractors tailored from histologyn such as https://github.com/owkin/HistoSSLscaling have been open sourced in recent years.

Predictive models training

The script to train models to predict the TEAD_500 signature in a cross-validated fashion is available here: he2tead/engine/run_cv.py.

An example is provided as a bash script

cd he2tead/engine
bash run.sh