This repository contains the first Use Case Application for UC#1 of the PROCESS project, http://www.process-project.eu/, in particular the second layer of the software architecture CamNet. The use case tackles cancer detection and tissue classification on the latest challenges in cancer research using histopathology images from CAMELYON 16 and 17. The software implemented by the use case consists of three layers. L 1. Data extraction and preprocessing: https://github.com/medgift/PROCESS_L1 L 2. Network training L 3. Network interpretability

The code is written in Python 2.7 and requires Keras 2.1.5 with Tensorflow 1.4.0 as backend. Further dependencies are in requirements.txt.

Configuration files are ini-based. A full template is in doc/config.cfg.

The master script is a pipeline-based program that can be run by the command

python train_cnn.py GPU_DEVICE EXPERIMENT_NAME RANDOM_SEED

GPU_DEVICE = GPU index on server EXPERIMENT_NAME = name of the experiment RANDOM_SEED = seed for reproducibility.

For example, you can run the script as follows:

python train_cnn.py 0 debug_run 1001

Network training is distributed on multiple GPUs by using the Horovod toolkit (https://github.com/horovod/horovod). To access the files relative to this modality you should checkout the horovod branch of this repository:

git checkout horovod