RACLAHE_Image_Enhancement_for_CNN_model_segmentation

RACLAHE: Region Adaptive MR Image Enhancement for CNN-based Segmentation

RACLAHE (Region-Adaptive Contrast Limited Adaptive Histogram Equalization) is an image enhancement method specifically designed for improving CNN-based segmentation of the prostate and prostatic zones in T2-Weighted MR images.

This repository supports the original publication in Nature’s Scientific Reports: “Region-Adaptive Magnetic Resonance Image Enhancement for improving CNN-based segmentation of the prostate and prostatic zones” doi:10.1038/s41598-023-27671-8

🚀 Quick Start

📖 See QUICKSTART.md for detailed step-by-step instructions

# 1. Prepare your data
mkdir -p input output shared
cp -r /path/to/your/patients/* input/

# 2. Run processing
docker-compose up --build

# 3. Get results from output/RACLAHE OUTPUT/

📋 Requirements

Docker and Docker Compose
Medical images in DICOM (.dcm) or NIfTI (.nii, .nii.gz) format

📁 Input Data Format

Organize your data with one folder per patient:

DICOM Series

input/
├── patient001/
│   ├── slice001.dcm
│   ├── slice002.dcm
│   └── ...
└── patient002/
    ├── slice001.dcm
    └── ...

NIfTI Files

input/
├── patient001/
│   └── scan.nii.gz
└── patient002/
    └── scan.nii.gz

📤 Output Format

Enhanced images are saved in the same format as input:

output/RACLAHE OUTPUT/
├── patient001/
│   ├── image_1.dcm      (if DICOM input)
│   ├── image_2.dcm
│   └── ...
└── patient002/
    └── patient002.nii.gz (if NIfTI input)

🔧 Usage

With Docker Compose (Recommended)

docker-compose up --build

With Docker

# Build
docker build -t raclage:3.0 .

# Run
docker run \
  -v $(pwd)/input:/home/ds/datasets:ro \
  -v $(pwd)/output:/home/ds/persistent-home \
  raclage:3.0

Custom Paths

The entrypoint accepts two optional positional arguments, INPUT_DIR and OUTPUT_DIR:

raclahe [INPUT_DIR] [OUTPUT_DIR] [--weights WEIGHTS_PATH]

Argument	Description	Default
`INPUT_DIR`	Dataset directory, one sub-directory per patient	`/home/ds/datasets`
`OUTPUT_DIR`	Where the enhanced images are written	`/home/ds/persistent-home/output`
`--weights`	Bounding-box U-Net weights (`.h5`)	`/home/ds/bbox_weights/checkpoint_external.h5`

docker run \
  -v /your/input:/custom/input:ro \
  -v /your/output:/custom/output \
  raclahe:3.0 /custom/input /custom/output

The equivalent environment variables INPUT_DIR, OUTPUT_DIR and WEIGHTS_PATH are still honoured; command-line arguments take precedence.

On the EUCAIM platform

jobman submit -i raclahe -- <INPUT_DIR> <OUTPUT_DIR>

jobman submit -i raclahe -- ~/datasets/87f3be56-4725-45c3-9baa-d338de530f73/ ~/persistent-home/results/

OUTPUT_DIR should be a path under persistent-home, which is shared between all desktops and jobs of the user; otherwise the results are lost when the job ends.

✅ EUCAIM Platform Compliance

This image is fully EUCAIM-compliant:

✅ Runs as non-root user ds (uid 1000, gid 1000)
✅ Uses standard EUCAIM volume mount points:
- /home/ds/datasets - Input datasets (read-only)
- /home/ds/persistent-home - Output storage
- /home/ds/persistent-shared-folder - Shared storage
✅ No internet access required at runtime
✅ Batch processing optimized for medical imaging workflows

🧬 Algorithm Overview

RACLAHE enhances medical images by:

Detecting the prostate region using a pre-trained U-Net model
Applying adaptive histogram equalization to the detected region
Preserving original characteristics in non-prostatic regions
Combining enhanced and unenhanced regions for the final output

This targeted approach improves CNN segmentation performance by 3-9% (Dice score) across different prostatic regions.

📊 Citation

If you use RACLAHE in your research, please cite:

@article{zaridis2023region,
  title={Region-adaptive magnetic resonance image enhancement for improving CNN-based segmentation of the prostate and prostatic zones},
  author={Zaridis, Dimitrios I and Mylona, Eugenia and Tachos, Nikolaos and Pezoulas, Vasileios C and Grigoriadis, Grigorios and Tsiknakis, Nikos and Marias, Kostas and Tsiknakis, Manolis and Fotiadis, Dimitrios I},
  journal={Scientific Reports},
  volume={13},
  number={1},
  pages={714},
  year={2023},
  publisher={Nature Publishing Group UK London}
}

🆘 Support

For questions or issues:

📖 Read the QUICKSTART.md guide
📧 Contact: dimzaridis@gmail.com
📄 Paper: https://doi.org/10.1038/s41598-023-27671-8

📜 License

Licensed under the European Union Public Licence (EUPL) v1.2 - see LICENSE and NOTICE.

🙏 Acknowledgements

This work is supported by the ProCancer-I project, funded by the European Union’s Horizon 2020 research and innovation program under grant agreement No 952159.

This site is open source. Improve this page.