RANO-classification

This repository contains the scripts used to classify the RANO criteria used in the following paper: Towards a deep learning approach for classifying treatment response in glioblastomas

It includes scripts for preprocessing and organizing of the data from the LUMIERE dataset.

Table of Contents

• How to use
  • 1. Alter CSV and python package
  • 2. Run Preprocessing and Organization
  • 3. Run Benchmarking Script
  • 4. Run Explainability Script

How to use

First, clone the repository using:

git clone https://github.com/anamatoso/RANO-classification.git

Copy the LUMIERE folder into this folder.

Then create a virtual environment, activate it, and install the requirements:

python -m venv venv
source ./venv/bin/activate
pip install -r requirements.txt

Check whether the packages were indeed installed using pip list. If not use python3 -m pip install -r requirements.txt.

1. Alter CSV and python package

In the LUMIERE-ExpertRating-v202211.csv file in line 172, line 578 and in line 613 delete the extra space after "Post-Op". Additionally, change the "Date" header to "Timepoint", the "Rating (according to RANO, PD: Progressive disease, SD: Stable disease, PR: Partial response, CR: Complete response, Pre-Op: Pre-Operative, Post-Op: Post-Operative)" to "RANO" and the "Rating rationale (CRET: complete resection of the enhancing tumor, PRET: partial resection of the enhancing tumor, T2-Progr.: T2-Progression, L: Lesion)" to "RatingRationale". The header should not have inverted commas.

Then, add the following transformation to the file ./venv/lib/python[VERSION]/site-packages/monai/transforms/utility/dictionary.py (replace [VERSION] with the one you're using, in my case it was 3.9) in line 926 and add its name (SubtractItemsd) to line 159. Additionally, add its name also to the file ./venv/lib/python[VERSION]/site-packages/monai/transforms/__init__.py in line 622 so that the package is aware of it.

class SubtractItemsd(MapTransform):
    """
    Subtract specified items from data dictionary elementwise.
    Expect all the items are numpy array or PyTorch Tensor or MetaTensor.
    Return the first input's meta information when items are MetaTensor.
    """

    backend = [TransformBackends.TORCH, TransformBackends.NUMPY]


    def __init__(self, keys: KeysCollection, name: str, allow_missing_keys: bool = False) -> None:
        """
        Args:
            keys: keys of the corresponding items to be subtracted.
                See also: :py:class:`monai.transforms.compose.MapTransform`
            name: the name corresponding to the key to store the resulting data.
            allow_missing_keys: don't raise exception if key is missing.
        """
        super().__init__(keys, allow_missing_keys)
        self.name = name

    def __call__(self, data: Mapping[Hashable, NdarrayOrTensor]) -> dict[Hashable, NdarrayOrTensor]:
        """
        Raises:
            TypeError: When items in ``data`` differ in type.
            TypeError: When the item type is not in ``Union[numpy.ndarray, torch.Tensor, MetaTensor]``.

        """
        d = dict(data)
        output = []
        data_type = None
        for key in self.key_iterator(d):
            if data_type is None:
                data_type = type(d[key])
            elif not isinstance(d[key], data_type):
                raise TypeError("All items in data must have the same type.")
            output.append(d[key])

        if len(output) == 0:
            return d

        if data_type is np.ndarray:
            d[self.name] = np.subtract(output[0], output[1])
        elif issubclass(data_type, torch.Tensor):  
            d[self.name] = torch.sub(output[0], output[1])
        else:
            raise TypeError(
                f"Unsupported data type: {data_type}, available options are (numpy.ndarray, torch.Tensor, MetaTensor)."
            )
        return d

2. Run Preprocessing and Organization

Run both the 01_preprocessing.py and the 02_organize_data.py:

python ./01_preprocessing.py
python ./02_organize_data.py

3. Run Benchmarking Script

Before running the 03_benchmarking.py file you must add a transform class to the monai package.

Then you can run an experiment. An example of an experiment to run would be:

python 03_benchmarking.py --model_name monai_densenet264 --n_epochs 100 --decrease_LR --mods_keep T1,T2,FLAIR

4. Run Explainability Script

The explainability script must be run as a jupyter notebook, with cells.

In the first run (which will be for class 0), run every cell except the "Select class" cell in order. Then for classes 1 until 3, run from the cell "Select class" onwards, one time for each class.