Related Experiment Video
Updated: Jun 30, 2025
08:41
Assessment of Cardiac Function and Myocardial Morphology Using Small Animal Look-locker Inversion Recovery SALLI MRI in Rats
Published on: July 19, 2013
12.8K
Recurrent neural network-based simultaneous cardiac T1, T2, and T1ρ mapping
Yiming Tao1, Zhenfeng Lv1, Wenjian Liu1
1School of Biomedical Engineering, ShanghaiTech University, Shanghai, China.
NMR in Biomedicine
|March 23, 2024
Summary
A deep neural network using a recurrent neural network (RNN) accelerates cardiac mapping by generating T1, T2, and T1ρ maps in 2 seconds. This method offers similar accuracy to conventional techniques but is 60x faster, enabling efficient clinical applications.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Cardiovascular MRI
Background:
- Multiparametric mapping provides comprehensive tissue characterization in cardiac MRI.
- Current free-breathing cardiac mapping techniques can be time-consuming, limiting clinical utility.
- Accelerating image reconstruction is crucial for efficient multiparametric cardiac MRI.
Purpose of the Study:
- To assess the feasibility of training a deep neural network (RNN) to accelerate the generation of T1, T2, and T1ρ maps.
- To develop an RNN-based model for rapid and accurate estimation of cardiac T1, T2, and T1ρ parameters.
- To compare the performance of the RNN-based method against dictionary-matching and conventional mapping techniques.
Main Methods:
- A recurrent neural network (RNN) was trained using a large dataset (>10 million signals) generated via Bloch simulations with varying noise levels.
- The RNN model exploited temporal correlations in multicontrast images for parameter estimation.
- Performance was evaluated in phantom studies and in vivo studies with 10 healthy volunteers at 3T, comparing against dictionary-matching and conventional methods.
Main Results:
- The RNN-based method and dictionary-matching method demonstrated comparable accuracy and precision in T1, T2, and T1ρ estimations in both phantom and in vivo studies.
- In vivo results showed no significant differences (p > 0.1) in T1, T2, and T1ρ values between the RNN and dictionary-matching methods.
- The RNN-accelerated method achieved simultaneous reconstruction of cardiac multiparameter maps in just 2 seconds, representing a 60-fold acceleration.
Conclusions:
- The RNN-based deep learning approach enables highly accelerated, accurate, and simultaneous reconstruction of T1, T2, and T1ρ maps for free-breathing cardiac multiparametric mapping.
- This method significantly improves efficiency compared to dictionary-matching techniques.
- The RNN-accelerated technique holds promise for enabling inline quantitative mapping in clinical cardiovascular MRI applications.
