This repository contains the code to run a tissue-background segmentation based on Resolution-agnostic tissue segmentation in whole-slide histopathology images with convolutional neural networks. It was used in the acrobat challenge to mask away the control tissue, see here.
The software is provided on "AS IS" basis, i.e. it comes without any warranty, express or implied including (without limitations) any warranty of merchantability and warranty of fitness for a particular purpose.
The network was trained on two GPUs, but works on CPU for inference. To call the inference code, you can build a docker image or a python package.
For a quick start, you can use the docker image. It outputs the biggest connected tissue segmentation, which was used to ignore the control tissue.
Go to minimal_inference_docker for more details on how to build and run the docker image.
For a more fine grained control, you can use the python package. It provides functionalities for tissue segmentation with and without post processing, which only chooses the biggest connected segmentation.
Go to tissue_segmentation for more details on how to build and run the python package.