This project contains an implementation of a framework for multi-class histopathology image segmentation written in Python 3.8. The deep learning backbone uses Tensorflow/Keras. We use Conda for managing the packages.
./segmentations/ is a stub; Because of the size of the images, we only attach segmentations sub-sampled at 1/4 of the resolution in a separate .zip file attached to the thesis.
./data/ contains the data set; Because of the size of the data set, we only include the validation images. The annotations can be viewed using ASAP (see below).
./histoseg/ contains the implementation of the project. You can also find there the Conda environment configuration file requirements.txt.
./histoseg/ml/notebooks/ contains Jupyter notebooks with implementation of the models. All the models are sub-classed from the ModelPipeline class.
./miscellaneous/mapping.png shows the color overlay we use for the individual classes.
./expert_evaluations/ contains the original expert's evaluation (in Czech) we use in the thesis.
The project uses Anaconda. To create a new virtual environment, use the following command.
conda env create -f histoseg/environment.yml
The name of the new environment will be 'histoseg'.
Additionally, a docker image of ASAP can be used to view the annotations. For installation, follow the instructions at https://hub.docker.com/r/vladpopovici/asap.
The basic useage is described in histoseg/notebooks/Example.ipynb
The data preparation procedure is described in histoseg/notebooks/DataExample.ipynb
For additional questions, contact me at tjelinek@mail.muni.cz