Bloch simulator-driven deep recurrent neural network for magnetization transfer contrast MR fingerprinting and CEST
Munendra Singh1, Shanshan Jiang1, Yuguo Li1
1Division of MR Research, Department of Radiology, Johns Hopkins University, Baltimore, Maryland, USA.
Magnetic Resonance in Medicine
|June 15, 2023
Summary
This study introduces a deep-learning framework for Magnetization Transfer Contrast (MTC) MR Fingerprinting (MRF) that significantly speeds up calculations and improves accuracy for tissue parameter quantification.
Area of Science:
- Magnetic Resonance Imaging
- Deep Learning
- Biophysics
Background:
- Magnetization Transfer Contrast (MTC) is crucial for characterizing tissue properties.
- MR Fingerprinting (MRF) offers efficient parameter mapping.
- Integrating MTC with MRF presents computational challenges.
Purpose of the Study:
- To develop a unified deep-learning framework for MTC-MRF.
- To enable accurate estimation of MTC effects using MRF.
- To combine an ultrafast Bloch simulator with MRF reconstruction.
Main Methods:
- Developed recurrent and convolutional neural network architectures for Bloch simulation and MRF reconstruction.
- Evaluated the framework using numerical phantoms and in vivo human brain scans at 3T.
- Assessed MTC-MRF, CEST, and NOE imaging, including test-retest repeatability.
Main Results:
- The deep Bloch simulator reduced computation time by 181-fold compared to conventional methods.
- The deep-learning MRF reconstruction demonstrated superior accuracy and noise robustness.
- Test-retest studies showed high repeatability for all quantified tissue parameters (CV < 7%).
Conclusions:
- Deep learning-driven MTC-MRF provides robust and repeatable multi-tissue parameter quantification.
- The framework achieves this within clinically feasible scan times on a 3T scanner.
- This approach enhances the utility of MR fingerprinting for MTC analysis.
