Related Experiment Video
Updated: Apr 28, 2026
08:45
Mapping Cortical Dynamics Using Simultaneous MEG/EEG and Anatomically-constrained Minimum-norm Estimates: an Auditory Attention Example
Published on: October 24, 2012
14.2K
DeepRelaxo: Fast Mono-Exponential Magnitude Brain R2* Mapping With Reduced Echoes Using Self-Supervised Deep
Samiha Prima1, Zhuang Xiong2, Alan H Wilman3
1School of Electrical Engineering and Computer Science, University of Queensland, Brisbane, Queensland, Australia.
Magnetic Resonance in Medicine
|April 27, 2026
Summary
DeepRelaxo, a new deep learning method, accurately estimates brain R2* maps from fast multi-echo gradient echo scans. This approach enhances imaging speed and robustness, even with reduced data acquisition.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Medical Physics
Background:
- Quantitative susceptibility mapping (QSM) relies on accurate R2* estimation.
- Multi-echo gradient echo (ME-GRE) sequences are crucial for R2* mapping.
- Accelerated imaging protocols are needed for clinical applications.
Purpose of the Study:
- Introduce DeepRelaxo, a deep learning method for fast and generalizable R2* mapping.
- Enable R2* estimation from ME-GRE acquisitions with arbitrary echo configurations.
- Support accelerated scans using shortened echo trains.
Main Methods:
- DeepRelaxo employs a two-stage self-supervised network: Transformer-MLP for initial R2* estimation and 3D U-Net for denoising.
- The network is trained on simulated multi-echo gradient echo data with varied parameters.
- Evaluated against non-linear least squares (NLLS) and Transformer-MLP on simulated and in vivo data.
Main Results:
- DeepRelaxo outperforms NLLS and Transformer-MLP in simulations, especially under accelerated conditions (e.g., 4x speedup).
- Achieved significant improvements in SSIM (13.5%) and RMSE (76%) at low SNR (SNR=10).
- In vivo data (3T and 7T) showed consistent R2* values in deep gray matter and preserved anatomical detail.
Conclusions:
- DeepRelaxo accurately models ME-GRE decay using temporal and spatial context.
- Provides robust and computationally efficient R2* mapping.
- Enables reliable reconstruction for accelerated protocols in time-sensitive workflows.
