Physics-Guided Self-Supervised Learning: Demonstration for Generalized RF Pulse Design

• Introduction
• Getting Started
  • Installation
  • Offline Training
  • Online Adaptation
• Publication
• Contacts

Introduction

Generalized RF pulse design using Physics-guided Self-supervised learning (GPS) is a comprehensive framework for creating MRI RF pulses using a physics-guided, self-supervised learning approach. This algorithm can generate RF pulses that meet flexible design requirements by utilizing a physics module, the Bloch equation, to guide the learning and optimization process. The pulse design for 1D selective pulse, B1-insensitive pulse, SPatial-SPectral (SPSP) pulse, and 2D pulse is demonstrated. Through online adaptation, GPS can further compensate for MRI system imperfections, such as field inhomogeneity. For more details, see our [paper] published on Magnetic Resonance in Medicine.

Getting Started

The computing environment we tested.

• CPU: Intel Xeon Gold 6338 @ 2.00GHz
• GPU: NVIDIA A100

Installation

0. Download and Install the appropriate version of NVIDIA driver and CUDA for your GPU.
1. Download and install Anaconda or Miniconda.
2. Clone this repo and cd to the project path.

git clone git@github.com:I3Tlab/GPS_RF.git
cd GPS_RF

3. Create and activate the Conda environment:

conda create --name GPSRF python=3.10.12
conda activate GPSRF

4. Install dependencies

pip install -r requirements.txt

Offline training

Offline training indicates RF pulse design with homogenous fields.

1D selective RF pulse (Figure 2a in the paper)

python 1D_pulse_demo.py

1D B1-insensitive RF pulse (Figure 2b in the paper)

python 1D_adiabatic_demo.py

SPSP RF pulse (Figure 4 in the paper)

Fat suppression (Figure 4a in the paper)

python 1D_SPSP_demo.py --data_path data_loader/SPSP_TBW_3_SBW_6_pw_23p8ms_exc_width_5mm_water_192x96_conj.mat --notes water

Fat saturation (Figure 4b in the paper)

python 1D_SPSP_demo.py --data_path data_loader/SPSP_TBW_3_SBW_6_pw_23p8ms_exc_width_5mm_fat_192x96_conj.mat --notes fat

2D RF pulse (Figure 5 in the paper)

python 2D_AI_demo.py

Online Adaptation

Online adaptation indicates field inhomogeneity compensation by adjusting the RF pulse.

online adaptation for phantom scan (Figure 7 in the paper)

python 2D_online_adaptation_demo.py --B0 data_loader/measured_B0_20240407_3_phantom.mat --B1 data_loader/measured_B1_20240407_phantom.mat --notes phantom

online adaptation for phantom invivo brain scan (Figure 8 in the paper)

python 2D_online_adaptation_demo.py --B0 data_loader/measured_B0_20240415_2_brain.mat --B1 data_loader/measured_B1_20240415_brain.mat --notes invivo_brain

Publication

@article{https://doi.org/10.1002/mrm.30307,
author = {Jang, Albert and He, Xingxin and Liu, Fang},
title = {Physics-guided self-supervised learning: Demonstration for generalized RF pulse design},
journal = {Magnetic Resonance in Medicine},
year = {2024}
volume = {early access},
number = {early access},
pages = {1-16},
keywords = {Bloch equations, deep learning, GPS, online adaptation, RF pulse, self-supervised learning},
doi = {https://doi.org/10.1002/mrm.30307},
url = {https://onlinelibrary.wiley.com/doi/abs/10.1002/mrm.30307},
}

Contacts

Intelligent Imaging Innovation and Translation Lab [github] at the Athinoula A. Martinos Center of Massachusetts General Hospital and Harvard Medical School

• Albert Jang (awjang@mgh.harvard.edu)
• Xingxin He (xihe2@mgh.harvard.edu)
• Fang Liu (fliu12@mgh.harvard.edu)

149 13th Street, Suite 2301 Charlestown, Massachusetts 02129, USA