Official pytorch implementation of Efficient Dual-line Decoder with Multi-Scale Convolutional Attention for Multi Organ segmentation which is published in Biomedical Signal Processing and Control Journal.

DOI: https://doi.org/10.1016/j.bspc.2025.108611

PDF: arxiv

Riad Hassan, M. Rubaiyat Hossain Mondal, Sheikh Iqbal Ahamed, Fahad Mostafa, Md Mostafijur Rahman

Fig: Overall architecture of EDLDNet

Fig: Dice score vs MACs count for different segmentation methods over synapse dataset.

Fig: The comparison of contoured segmentation images from Synapse dataset among proposed method and the competitive existing methods.

The project is implement with Python 3.8 and pytorch 1.11.0+cu113. To install pytorch you can use:

pip install torch==1.11.0+cu113 torchvision==0.12.0+cu113 torchaudio==0.11.0 --extra-index-url https://download.pytorch.org/whl/cu113

Now install the required library using following command:

pip install -r requirements.txt

You should download the pretrained PVTv2 model from Google Drive / PVT GitHub, and then put it in the ./pretrained_pth/pvt/ folder for initialization.

• Pre-Processed Synapse Multi-organ dataset: Sign up in the official Synapse website and download the dataset. Then split the 'RawData' folder into 'TrainSet' (18 scans) and 'TestSet' (12 scans) following the TransUNet's lists and put in the './data/synapse/Abdomen/RawData/' folder. Finally, preprocess using python ./utils/preprocess_synapse_data.py. OR Download the preprocessed data and save in the ./data/synapse/ folder.

python -W ignore train_synapse.py --root_path /path/to/train/data --volume_path path/to/test/data         # replace --root_path and --volume_path arguments with your actual path

Trained Model Weight:

Download the weight or use your trained model weight to test

python -W ignore test_synapse.py --saved_model_path /path/to/best/model/weight      # replace --saved_model_path argument with actual path

After completion of testing, check the test_log directory to get the test result.

We sincerely acknowledge and deeply appreciate the outstanding contributions of timm, TransUNet, EMCAD and UDBANet, whose remarkable works have laid the foundation for the development of our framework.

@article{HASSAN2026108611,
    title = {An efficient dual-line decoder network with multi-scale convolutional attention for multi-organ segmentation},
    journal = {Biomedical Signal Processing and Control},
    volume = {112},
    pages = {108611},
    year = {2026},
    issn = {1746-8094},
    doi = {https://doi.org/10.1016/j.bspc.2025.108611},
    url = {https://www.sciencedirect.com/science/article/pii/S174680942501122X},
    author = {Riad Hassan and M. Rubaiyat Hossain Mondal and Sheikh Iqbal Ahamed and Fahad Mostafa and Md Mostafijur Rahman}
}