Fuzzy-Rank-Ensemble-Torch

A PyTorch-based implementation of a Fuzzy Rank-based Ensemble compatible with the MONAI framework for medical image segmentation and classification tasks.

The code is based on our papers:

• "A fuzzy rank-based ensemble of CNN models for MRI segmentation" published in Elseiver Biomedical Signal Processing and Control journal.
• "A Fuzzy Rank-based Ensemble of CNN Models for Classification of Cervical Cytology" published in Nature-Scientific Reports journal.

Installation

Basic Installation

pip install fuzzy-rank-ensemble-torch

With MONAI Support

pip install fuzzy-rank-ensemble-torch[monai]

Development Installation

git clone https://github.com/Digiratory/fuzzy-rank-ensemble-torch.git
cd fuzzy-rank-ensemble-torch
pip install -e .

Usage

Pure PyTorch Implementation

import torch
from fuzzy_rank_ensemble_torch import fuzzy_rank_ensemble

# Initialize prediction tensors from different models
# Shape: [Batch, Class Index]

# Inception v3 model predictions
inception_v3_pred = torch.tensor([[0.261, 0.315, 0.102, 0.286]])

# Xception model predictions
xception_pred = torch.tensor([[0.402, 0.347, 0.201, 0.050]])

# DenseNet-169 model predictions
densenet_169_pred = torch.tensor([[0.357, 0.467, 0.131, 0.045]])

# Apply fuzzy rank ensemble
result = fuzzy_rank_ensemble([inception_v3_pred, xception_pred, densenet_169_pred])
print(f"Ensemble Result: {result}")

MONAI Integration

import torch
from fuzzy_rank_ensemble_torch.monai import FuzzyRankBasedEnsemble

# Initialize models and predictions (same as above)
# ...

# Create ensemble instance
ens = FuzzyRankBasedEnsemble()

# Apply ensemble to predictions
result = ens([inception_v3_pred, xception_pred, densenet_169_pred])

# Verify result matches expected output
expected = torch.tensor([[1-0.4587, 1-0.4264, 1-0.5948, 1-0.5883]])
assert torch.allclose(result, expected, rtol=1e-03)

Also, you can use dictionary-based wrapper FuzzyRankBasedEnsembled.

Citation

If you use this software in your research, please cite our work:

@article{10.1016/j.bspc.2024.107342,
  title     = {A fuzzy rank-based ensemble of CNN models for MRI segmentation},
  journal   = {Biomedical Signal Processing and Control},
  volume    = {102},
  pages     = {107342},
  year      = {2025},
  issn      = {1746-8094},
  doi       = {10.1016/j.bspc.2024.107342},
  url       = {https://www.sciencedirect.com/science/article/pii/S1746809424014009},
  author    = {Valenkova, Daria and Lyanova, Asya and Sinitca, Aleksandr and Sarkar, Ram and Kaplun, Dmitrii},
}

@article{manna2021fuzzy,
  title        = {A fuzzy rank-based ensemble of CNN models for classification of cervical cytology},
  author       = {Manna, Ankur and Kundu, Rohit and Kaplun, Dmitrii and Sinitca, Aleksandr and Sarkar, Ram},
  journal      = {Scientific Reports},
  volume       = {11},
  number       = {1},
  pages        = {1--18},
  year         = {2021},
  doi          = {10.1038/s41598-021-93783-8},
  publisher    = {Nature Publishing Group}
}

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

License

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