Prostate cancer segmentation using multiparamteric MRI from Pi-CAI challenge dataset

Prerequisites

The software is developed in Python 3.10. For deep learning, the PyTorch 1.13 framework is used.

Main Python modules required for the software can be installed from ./requirements in three stages:

1. Create a Python3 environment by installing the conda environment.yml file:

$ conda env create -f environment.yaml
$ source activate segpicai

2. Install the remaining dependencies from requirements.txt.

Note: These might take a few minutes.

Code structure

Our source code for training and evaluation of the deep neural networks, image analysis and preprocessing, and data augmentation are available here.

1. Everything can be run from ./main_3D_picai.py.

• The data preprocessing parameters, directories, hyper-parameters, and model parameters can be modified from ./configs/config.yaml.
• Also, you should first choose an experiment name (if you are starting a new experiment) for training, in which all the evaluation and loss value statistics, tensorboard events, and model & checkpoints will be stored. Furthermore, a config.yaml file will be created for each experiment storing all the information needed.
• For testing, just load the experiment which its model you need.

2. The rest of the files:

• ./data/ directory contains all the data preprocessing, augmentation, and loading files.
• ./Train_Valid_picai.py contains the training and validation processes.
• ./Prediction_picai.py all the prediction and testing processes.