DP-HAFNet

DP-HAFNet is a novel deep learning framework for high-quality ultrasound image reconstruction from single plane-wave radio frequency (RF) data. This network features a dual-path architecture that simultaneously predicts pixel-level adaptive weights and channel-weighted beamforming outputs, enabling superior image quality while maintaining computational efficiency.

📄 Corresponding Publication

This repository contains the implementation of the method described in our paper:

DP-HAFNet: A dual-path hierarchical adaptive fusion network for ultrasound image reconstruction from single plane-wave RF data

🛠 Environment Setup

This project requires Python 3.8 and PyTorch 2.0.1. Follow these steps to set up the environment:

1. Clone Repository

git clone https://github.com/Lijiajin0719/DP-HAFNet.git
cd DP-HAFNet

2. Create Conda Environment

conda create -n dp-hafnet python=3.8
conda activate dp-hafnet

3. Install Dependencies

pip install -r requirements.txt

📊 Dataset

1.Download datasets from CUBDL and PICMUS

2.Extract files to the data directory

If you use this dataset, please cite the following article:

CUBDL

@INPROCEEDINGS{9251434,
  author={Bell, Muyinatu A. Lediju and Huang, Jiaqi and Hyun, Dongwoon and Eldar, Yonina C. and van Sloun, Ruud and Mischi, Massimo},
  booktitle={2020 IEEE International Ultrasonics Symposium (IUS)}, 
  title={Challenge on Ultrasound Beamforming with Deep Learning (CUBDL)}, 
  year={2020},
  pages={1-5},
  keywords={Deep learning;Measurement;Image quality;Ultrasonic imaging;Array signal processing;Benchmark testing;Task analysis},
  doi={10.1109/IUS46767.2020.9251434}
}

PICMUS

@INPROCEEDINGS{7728908,
  author={Liebgott, H. and Rodriguez-Molares, A. and Cervenansky, F. and Jensen, J.A. and Bernard, O.},
  booktitle={2016 IEEE International Ultrasonics Symposium (IUS)}, 
  title={Plane-Wave Imaging Challenge in Medical Ultrasound}, 
  year={2016},
  pages={1-4},
  keywords={Ultrasonic imaging;Measurement;Biomedical imaging;Image resolution;Speckle;Array signal processing;ultrasound;challenge;Plane-Wave;beamforming;ultrafast},
  doi={10.1109/ULTSYM.2016.7728908}
}

🏋️ DP-HAFNet train/test

1. Train

python train.py --num_epochs 2000 --batch_size 8 --gpu_id 0 --log_dir logs

To see more intermediate results, check out ./logs/DP_HAFNet....

2. Test

python test.py --model_path logs/... --gpu_id 0 --log_dir logs --save_images --save_mat

The test results will be saved to file here: ./logs/test....

📧 Contact

For any questions regarding the paper or this implementation, please feel free to contact the authors.

📩 Email: 15563866837@163.com

📝 Citation

If you use this code for your research, please cite our paper.

@article{LI2026121562,
title = {DP-HAFNet: A dual-path hierarchical adaptive fusion network for ultrasound image reconstruction from single plane-wave RF data},
journal = {Measurement},
volume = {277}
pages = {121562},
year = {2026},
issn = {0263-2241},
doi = {https://doi.org/10.1016/j.measurement.2026.121562},
url = {https://www.sciencedirect.com/science/article/pii/S0263224126012716},
author = {Jiajin Li and Wenwen Sun and Yuanguo Wang and Chichao Zheng and Zhihui Han and Hu Peng and Chaoxue Zhang and Jinhua Zhou and Yadan Wang},
keywords = {Ultrasound image reconstruction, Plane-wave, Adaptive beamforming, Deep learning, Linear attention mechanism}
}

---

🌟 We appreciate your interest in our work!