Skip to content
/ ICARE Public

Individual Coefficient Approximation for Risk Estimation (ICARE) model

License

Notifications You must be signed in to change notification settings

Lrebaud/ICARE

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

61 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

PyPI version Downloads Python package codecov License Python 3.6


ICARE

Rebaud, L., Escobar, T., Khalid, F., Girum, K., Buvat, I. (2023). Simplicity Is All You Need: Out-of-the-Box nnUNet Followed by Binary-Weighted Radiomic Model for Segmentation and Outcome Prediction in Head and Neck PET/CT. In: Andrearczyk, V., Oreiller, V., Hatt, M., Depeursinge, A. (eds) Head and Neck Tumor Segmentation and Outcome Prediction. HECKTOR 2022. Lecture Notes in Computer Science, vol 13626. Springer, Cham.

This package contains the Individual Coefficient Approximation for Risk Estimation (ICARE) survival model. For ensembling strategies, it also includes a dedicated bagging aggregator.

This model was the winning solution for the MICCAI 2022 challenge: HEad and neCK TumOR (HECKTOR) for the outcome prediction task from PET/CT.

Description

The Individual Coefficient Approximation for Risk Estimation (ICARE) model uses a minimal learning strategy to reduce to risk of overfitting. To do so:

  • drop highly correlated features
  • for each feature:
    • evaluate feature sign using an univariate approach
    • normalize the feature
    • multiply the feature by its sign
  • the prediction is computed as the mean of all signed features

This makes the model more robust to overfitting. It also makes it resilient to the curse of dimensionality. We hypothesize that it is better to have too many features than too few for this model. This algorithm is implemented in the following classes:

  • IcareClassifier estimator for classification tasks
  • IcareRanker estimator for ranking tasks (not calibrated regression)
  • IcareSurvival estimator for survival prediction tasks package.

To improve the performance, this package also provides bagged versions of these estimators:

  • BaggedIcareClassifier estimator for classification tasks
  • BaggedIcareRanker estimator for ranking tasks (not calibrated regression)
  • BaggedIcareSurvival estimator for survival prediction tasks package.

The survival models (IcareSurvival and BaggedIcareSurvival) predict a risk score. Therefore, there predictions are anti-concordants with the target.

On the other hand, the ranking models (IcareRanker and BaggedIcareRanker) try to correctly the samples according to the target, so there predictions are concordants with the target.

Getting Started

Dependencies

  • Python 3.6 or later
  • pandas
  • seaborn
  • scikit-learn
  • scikit-survival

Installing

Via PyPI:

pip install icare

or via GitHub

pip install git+https://github.com/Lrebaud/ICARE.git

Documentation

Coming soon.

Utilisation

The model is used as any other scikit-learn estimator.

You can find detailed notebooks in the notebooks folder of this repository showing how to use the package for each type of datasets.

Author

Louis Rebaud: louis.rebaud@gmail.com

Version History

  • 0.1.0

    • Add classification and ranking support
  • 0.0.1

    • Initial Release

License

This project is licensed under the Apache License 2.0 - see the LICENSE.md file for details

Acknowledgements

This package was created as a part of the HEad and neCK TumOR segmentation and outcome prediction in PET/CT images challenge, 3rd edition (HECKTOR 2022) and presented by Louis Rebaud and Thibault Escobar at the 25th international conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) congress in Singapore.

Citation

If you use this package for your research, please cite the following paper:

Rebaud, L., Escobar, T., Khalid, F., Girum, K., Buvat, I. (2023). Simplicity Is All You Need: Out-of-the-Box nnUNet Followed by Binary-Weighted Radiomic Model for Segmentation and Outcome Prediction in Head and Neck PET/CT. In: Andrearczyk, V., Oreiller, V., Hatt, M., Depeursinge, A. (eds) Head and Neck Tumor Segmentation and Outcome Prediction. HECKTOR 2022. Lecture Notes in Computer Science, vol 13626. Springer, Cham.

About

Individual Coefficient Approximation for Risk Estimation (ICARE) model

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published