Ultrasound Lung Aeration Map via Physics-Aware Neural Operators

Overview

This repository contains the author's implementation of LUNA (Lung Ultrasound Neural operator for Aeration maps) proposed in "Ultrasound Lung Aeration Map via Physics-Aware Neural Operators", an approach for reconstructing lung aeration maps from ultrasound radio-frequency (RF) data using neural operators.

Installation

Prerequisites

• Python 3.8+
• MATLAB R2020a+ (for data generation and baseline methods only)
• CUDA-compatible GPU (recommended)

Python Dependencies

Install the required Python packages, or install from the requirements file:

python3 -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -r requirements.txt

Repository Structure

luna/
├── data/            # synthetic data generation and simulation (MATLAB + Python helpers)
├── luna_model/      # ML model training and inference
├── requirements.txt # Python dependencies
└── README.md

Quick Start

1. For luna ML model: Download required files in the Google drive and put under luna_model.

For synthetic data generation, download required files in the Google drive and put under data.

2. Data Generation and Simulation Please refer to the steps in Data generation README.

3. Train Luna

cd luna_model
python main.py

4. Run Inference

cd luna_model
# Set EVALUATE=True in main.py or pass an experiment name for wandb
python main.py eval_run

Notes

• We provide a CNN-based model as a baseline in resluna.py.
• Training expects raw_mean.pt and raw_std.pt in luna_model/ and dataset JSONs lung_train.json/lung_test.json.
• Data generation workflow and HPC scripts are documented in data/README.md.

Citation

If you find this work useful, please cite our paper:

@article{wang2025ultrasound,
  title={Ultrasound lung aeration map via physics-aware neural operators},
  author={Wang, Jiayun and Ostras, Oleksii and Sode, Masashi and Tolooshams, Bahareh and Li, Zongyi and Azizzadenesheli, Kamyar and Pinton, Gianmarco F and Anandkumar, Anima},
  journal={ArXiv},
  pages={arXiv--2501},
  year={2025}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.