HECKTOR 2021 challenge
This repository gives an example on how to preprocess the data of the HECKTOR challenge. Any other preprocessing is welcomed and any framework can be used for the challenge, the only requirement is to submit the results in the same coordinate system as the original CT images (same spacing and same origin). This repository also contains the code used to prepare the data of the challenge (DICOM to NIFTI, SUV computation and bounding box generation, not needed for the participants). Moreover, it contains an example of implementation to resample the data within the bounding boxes and resample back to the original resolution.
Download Data
To access the data, visit the challenge website: https://www.aicrowd.com/challenges/miccai-2021-hecktor and follow the instructions. The code included here was intended to work with a specific repository structure described in Section Project Organization. Following git clone https://github.com/voreille/hecktor.git, create a data/ folder in the repository and place the unzipped data in it.
Install Dependencies
To install the necessary dependencies you can use pip install -r requirements.txt. It is advised to use it within a python3 virtual environment.
Resample Data
Run python src/resampling/resample.py to crop and resample the data following the repository structure or use arguments (type python src/resamping/resample.py --help for more informations).
Evaluation
An example of how the segmentation (task 1) will be evaluated is illustrated in the notebook notebooks/evaluate_segmentation.ipynb. Note that the Hausdorff distance at 95 % implemented in https://github.com/deepmind/surface-distance will be used in the challenge (not the one found in src/evaluation/scores.py).
The concordance index used to evaluate task 2 and 3 is implemented in the function concordance_index(event_times, predicted_scores, event_observed=None) from the file src/aicrowd_evaluator/survival_metrics.py. It was adapted from https://github.com/CamDavidsonPilon/lifelines/blob/master/lifelines/utils/concordance.py to account for missing predictions (missing predictions are handled as non-concordant).
Submission
Dummy examples of correct submission for task 1 and 2 can be found in notebooks/example_seg_submission.ipynb and notebooks/example_surv_submission.ipynbrespectively.
Project Organization
├── README.md
├── data <- NOT in the version control
│ ├── resampled <- The data in NIFTI resampled and cropped according to the bounding boxes (bbox.csv).
│ ├── hecktor_nii <- The data converted in the nifty format with the original geometric frame,
| | e.i. the one downloaded form AIcrowd
│ └── bbox.csv <- The bounding box for each patient computed with bbox_auto function from src.data.bounding_box
├── requirements.txt <- The requirements file for reproducing the analysis environment, e.g.
│ generated with `pip freeze > requirements.txt`
├── Makefile <- Used to do set up the environment and make the conversion of DICOM to NIFTI
├── notebooks
| ├── example_seg_submission.ipynb <- Example of a correct submission for the segmentation task (task 1).
| ├── example_surv_submission.ipynb <- Example of a correct submission for the survival task (task 2).
│ └── evaluate_segmentation.ipynb <- Example of how the evaluation will be computed.
└── src <- Source code for use in this project
├── aicrowd_evaluator <- Source code for the evaluation on the AIcrowd platform
│ ├── __init__.py
│ ├── surface-distance/ <- code to compute the robust Hausdorff distance availabe at https://github.com/deepmind/surface-distance
│ ├── evaluator.py <- Define the evaluator class for task 1 and 2
│ ├── segmentation_metrics.py <- Define the metrics used in the segmentation task.
| ├── requirements.txt <- The requirements file specific to this submodule
│ └── survival_metrics.py <- Define the metrics used for the survival task.
├── data <- Scripts to generate data
│ ├── __init__.py
│ ├── bounding_box.py
│ ├── utils.py <- Define functions used in make_dataset.py
│ └── make_dataset.py <- Conversion of the DICOM to NIFTI and computation of the bounding boxes
├── evaluation
| ├── __init__.py
│ └── scores.py <- (DEPRECATED) used to illustrate how the segmentation is evaluated. Refer to `src/aicrowd_evaluator`
| submodule for the actual evaluation of the challenge.
└── resampling <- Code to resample the data
├── __init__.py
└── resample.py
Project based on the cookiecutter data science project template. #cookiecutterdatascience
8 Nov 2, 2022
1 Feb 16, 2022
86 Dec 31, 2022
11 Mar 20, 2022
15 Nov 3, 2022
106 Dec 14, 2022
23 Mar 9, 2022
97 Jan 7, 2023
2 Oct 30, 2022
697 Jan 5, 2023
104 Oct 26, 2022
23 Jan 25, 2022
13 Jan 3, 2022
18 May 4, 2022
128 Dec 8, 2022
82 Dec 29, 2022
91 Dec 21, 2022
106 Dec 20, 2022
218 Dec 27, 2022