---
license: mit
library_name: pytorch
pipeline_tag: image-segmentation
tags:
- medical-image-segmentation
- 3d-segmentation
- vessel-segmentation
- nnunet
- pytorch
---

# HiLo: Spatial-Spectral Hybrid High-Low Frequency Activation for Heart and Brain Vessel Segmentation

HiLo is a 3D vessel segmentation framework for cardio-cerebrovascular imaging. It jointly models spatial and spectral high-/low-frequency cues to preserve fine tubular details while encoding large-scale anatomical structure.

> This work is published in [Expert Systems with Applications](https://www.sciencedirect.com/science/article/pii/S0957417426023626).

[![github](https://img.shields.io/badge/GitHub-deepang--ai/HiLo-blue)](https://github.com/deepang-ai/HiLo) 
[![huggingface weights](https://img.shields.io/badge/%F0%9F%A4%97%20Weights-deepang/HiLo-yellow)](https://huggingface.co/deepang/HiLo) 

## Weights

- [ImageCAS checkpoint](https://huggingface.co/deepang/HiLo/blob/main/ImageCAS_checkpoint_best.pth)
- [CAS2023 checkpoint](https://huggingface.co/deepang/HiLo/blob/main/CAS2023_checkpoint_best.pth)

## Usage

Please use these weights with the official code repository:

```bash
git clone https://github.com/deepang-ai/HiLo.git
cd HiLo
```

Follow the installation, data preparation, training, evaluation, and inference instructions in the [HiLo README](https://github.com/deepang-ai/HiLo#readme). HiLo depends on the bundled project-adapted `nnUNet/` package in the code repository, so install it with:

```bash
pip install -e ./nnUNet
```

## Citation

If you use HiLo or these pretrained weights, please cite:

```bibtex
@article{HUANG2026133453,
title = {HiLo: Spatial-Spectral Hybrid High-Low Frequency Activation for Heart and Brain Vessel Segmentation},
journal = {Expert Systems with Applications},
pages = {133453},
year = {2026},
issn = {0957-4174},
doi = {https://doi.org/10.1016/j.eswa.2026.133453},
url = {https://www.sciencedirect.com/science/article/pii/S0957417426023626},
author = {Jiahui Huang and Xin Lei and Qiong Wang and Valentin Sinitsyn and Yun Zhu and Ying Hu and Hao Chen and Yan Pang}
}
```

Please also cite nnU-Net when using the bundled training framework:

```bibtex
@article{isensee2021nnunet,
  title={nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation},
  author={Isensee, Fabian and Jaeger, Paul F and Kohl, Simon AA and Petersen, Jens and Maier-Hein, Klaus H},
  journal={Nature Methods},
  volume={18},
  number={2},
  pages={203--211},
  year={2021},
  publisher={Nature Publishing Group}
}
```