Official Implementation of "Implicit U-KAN2.0: Dynamic, Efficient and Interpretable Medical Image Segmentation"[MICCAI 2025]

This is the official implementation of "Implicit U-KAN2.0: Dynamic, Efficient and Interpretable Medical Image Segmentation", accepted by MICCAI 2025.

Installation

Follow these steps to set up the environment:

1. Clone the repository:

   git clone git@github.com:robotMonkeyButler/implicit-conukan.git
   cd conMultUKAN

2. Create and activate Conda environment

   conda create -n conmultukan python=3.10
   conda activate conmultukan
   pip install -r requirement.txt

3. Verify the environment by running

   python -c "import torch; print(torch.__verison__)"

Datasets

Download the datasets and put them under the repo folder.

Usage

Training

To train a model, use the following command

python train.py --dataset <DATASET> --net <NETWORK> --batch-size <BATCH_SIZE> --epochs <EPOCHS> --lr <LEARNING_RATE>

Description:

--epochs: Number of training epochs (default: 500).

--batch-size: Batch size for training (default: 8).

--dataset: Dataset to use ( polyp, nuclei, breast_cancer, melanoma_segmentation).

--lr: Learning rate (default: 1e-4).

--gradient-accumulation: Mini-batch size for gradient accumulation (default: 1).

--pretrained: Boolean to load a pretrained model (default: False).

--checkpoint: Path to a checkpoint file (if --pretrained is True).

--solver: Solver of ODE block.

Testing

To test a model, use the following command

python test_2d.py --dataset <DATASET> --net <NETWORK> --path ./outputs --identifier <MODEL_IDENTIFIER>

Description:

--dataset: Dataset to use ( polyp, nuclei, breast_cancer, melanoma_segmentation).

--path: output path

--pretrained: Boolean to load a pretrained model (default: False).

--solver: Solver of ODE block.

--identifier: Identifier of training model

e.g.

python test_2d --dataset polyp --net conMultUKAN --path ./outputs --identifier conMultUKAN-rk4-2024-12-09T041730Z_best

Output Directory

The results of training and validation are saved in the ./outputs/ directory uner the respective dataset name. The structure is as follows:

./outputs/
   <dataset_name>/
      checkpoints/
         {identifier}.pt
      log_files/

Citation

@inproceedings{cheng2025implicit,
  title={Implicit U-KAN2. 0: Dynamic, efficient and interpretable medical image segmentation},
  author={Cheng, Chun-Wun and Zhao, Yining and Cheng, Yanqi and Montoya-Zegarra, Javier A and Sch{\"o}nlieb, Carola-Bibiane and Aviles-Rivero, Angelica I},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={304--314},
  year={2025},
  organization={Springer}
}