HiProtoNet

Official code for: HiProtoNet: Hyperbolic Hierarchy-aware Part Prototypes for Aortic Stenosis Severity Classification MICCAI 2025 ASMUS Workshop

Quick Start

1. Clone the repository

   git clone https://github.com/DeepRCL/HiProtoNet.git
   cd HiProtoNet

2. Install dependencies

   pip install --upgrade pip
   pip install torch torchvision torchaudio
   pip install moviepy==1.0.3 pandas wandb tqdm seaborn torch-summary opencv-python jupyter jupyterlab tensorboard tensorboardX imageio array2gif scikit-image scikit-learn torchmetrics termplotlib
   pip install --upgrade plotly einops transformers timm
   pip install git+https://github.com/geoopt/geoopt.git
   pip install --upgrade wandb
   pip install -e .

3. Prepare your data

   • Place your data in the data/ folder.
   • For custom datasets, refer to src/data/ for dataloader examples.

Training & Testing

Run training or evaluation from the project root:

python main.py --config_path="src/configs/<config-name>.yml" --run_name="<your_run_name>" --save_dir="logs/<your_run_name>"

Common options:

• --eval_only=True : Evaluate a trained model
• --eval_data_type="val" or "test" : Validation or test set
• --push_only=True : Project prototypes to nearest features

Example:

python main.py --config_path="src/configs/Hyperbolic_XProtoNet.yml" --run_name="HiProtoNet_test" --save_dir="logs/HiProtoNet/test_run"

Acknowledgements

Parts of the codebase related to Lorentzian geometry (e.g., src/utils/lorentz.py) are adapted from the MERU project:

• MERU: Hyperbolic Image-Text Representations (ICML 2023)
• Code: https://github.com/facebookresearch/meru

Developer Notes

HiProtoNet builds on ProtoASNet (MICCAI 2023). Most code is inherited; new features are in files with Hyperbolic or Lorentz in their names:

Key scripts for hyperbolic/Lorentz models:

• src/agents/Hyperbolic_XProtoNet.py, Hyperbolic_XProtoNet_e2e.py, Hyperbolic_Video_XProtonet_e2e.py
• src/models/Hyper_XProtoNet.py, Hyper_Video_XProtoNet.py
• src/utils/lorentz.py

Other files follow ProtoASNet conventions. For more details on ProtoASNet, see the DeepRCL fork: https://github.com/DeepRCL/ProtoASNet

Citation

If you use this code, please cite our paper:

@InProceedings{10.1007/978-3-032-06329-8_19,
	author    = {Vaseli, Hooman and Wu, Victoria and Kim, Diane and Tsang, Michael Y. and Gu, Ang Nan and Luong, Christina and Abolmaesumi, Purang and Tsang, Teresa S. M.},
	editor    = {Ni, Dong and Noble, Alison and Huang, Ruobing and Xue, Wufeng},
	title     = {HiProtoNet: Hyperbolic Hierarchy-Aware Part Prototypes for Aortic Stenosis Severity Classification},
	booktitle = {Simplifying Medical Ultrasound},
	year      = {2026},
	publisher = {Springer Nature Switzerland},
	address   = {Cham},
	pages     = {197--207},
	isbn      = {978-3-032-06329-8}
}