Automatic delineation of tumorigenic areas

The following instructions are for UBUNTU-based systems. Start by creating a python environment and install the dependencies. Recomended is to use the following:

user@user:~/current_directory$ cd astrachallege/Code
user@user:~/astrachallnege/Code$ conda create -n env-name python=3.10
user@user:~/astrachallnege/Code$ conda activate env-name
user@user:~/astrachallnege/Code$ pip install -r requirements.txt --no-cache

WARNING! The environment for training and testing was created with CUDA 12.1 installed. If your systems uses different versions of CUDA, please update --extra-index-url in requirements.txt accordingly to your CUDA version. The lists of index-urls for Pytorch with given CUDA version can be found here: https://pytorch.org/get-started/previous-versions/ .

To display needed arguments the main arguments

user@user:~/astrachallnege/Code$ python tumorigenesis.py --help
Automatic detection of tumorigenic ares
 [-h] {compute,train} ...

positional arguments:
  {compute,train}
    compute        Use this argument to segment tumorigenic areas with a trained AI model
    train          Use this argument to train an AI model to detect tumorigenic areas in MRI(s)

options:
  -h, --help       show this help message and exit

Training mode

In a bash terminal, in the correct directory type the following to get information about the training mode:

user@user:~/astrachallnege/Code$ python tumorigenesis.py train --help
usage: Automatic detection of tumorigenic areas train [-h] [--configuration CONFIGURATION] [--mode MODE]

options:
  -h, --help            show this help message and exit
  --configuration CONFIGURATION
                        Provide the path of the 'config.yaml' file with the training specifications

To train a model from scratch

In the configuration file, you will find information about the parameters to train a new model from scratch. Have in mind that is better to have access to GPUs, otherwise the training will take significantly longer to converge. If the config.yaml file is located in the same directory as the rest of the code, to train a new model it is enough to type the following command in the terminal. Note that by doing this, you will use the same arguments as the submitted model.

user@user:~/astrachallnege/Code$ python tumorigenesis.py train

Computing mode

In a bash terminal, in the correct directory type the following to get information about the computing mode:

user@user:~/astrachallnege/Code$ python tumorigenesis.py compute --help
usage: Automatic detection of tumorigenic ares compute [-h] [--threshold THRESHOLD] [--device {cpu,cuda}] [--mode MODE] test_folder model_path model_name to_save

positional arguments:
  test_folder           Provide the directory storing the MRI(s)
  model_path            Provide the direcory storing the trained model
  model_name            Provide aproxy for the name of the model file
  to_save               Indicate the directory to save the predicted tumorigenic regions

options:
  -h, --help            show this help message and exit
  --threshold THRESHOLD
                        Indicate the threshold to binarize the probability maps
  --device {cpu,cuda}   Run inference on CPU or GPU

To compute segmentation masks using a trained model

python tumorigenesis.py compute testing-data-directory trained-model-location UNET results-directory

szmazurek / astrachallenge