Related Experiment Video
Updated: Jun 16, 2026
10:30
Analysis of Tubular Membrane Networks in Cardiac Myocytes from Atria and Ventricles
Published on: October 15, 2014
20.6K
DeepFittingNet: A deep neural network-based approach for simplifying cardiac T1 and T2 estimation with improved
Rui Guo1, Dongyue Si2, Yingwei Fan1
1School of Medical Technology, Beijing Institute of Technology, Beijing, China.
Magnetic Resonance in Medicine
|July 7, 2023
Summary
A new deep neural network, DeepFittingNet, simplifies cardiovascular MR mapping by accurately estimating T1/T2 values. It offers improved robustness for T1 estimation compared to traditional methods, enhancing data processing in cardiac imaging.
Area of Science:
- Cardiovascular Magnetic Resonance (CMR) imaging
- Artificial Intelligence in Medical Imaging
- Quantitative Cardiovascular Imaging
Background:
- Cardiovascular MR (CMR) mapping sequences like MOLLI, SASHA, and T2-prep bSSFP are crucial for quantitative tissue characterization.
- Traditional data processing for T1/T2 estimation relies on curve-fitting algorithms, which can be time-consuming and sensitive to noise.
- Developing automated and robust methods for T1/T2 estimation is essential for routine clinical application and research.
Purpose of the Study:
- To develop and evaluate DeepFittingNet, a deep neural network for T1/T2 estimation in common cardiovascular MR mapping sequences.
- To simplify data processing and enhance the robustness of T1/T2 quantification in CMR.
- To compare the performance of DeepFittingNet against conventional curve-fitting methods.
Main Methods:
- A 1D neural network (DeepFittingNet) comprising a recurrent neural network (RNN) and a fully connected neural network (FCNN) was designed.
- The network was trained using Bloch-equation simulations for MOLLI, SASHA, and T2-prep bSSFP sequences, with curve-fitting as the reference.
- Robustness was enhanced by simulating various imaging confounders, and the network was tested on phantom and in-vivo data.
Main Results:
- DeepFittingNet demonstrated robust T1/T2 estimation across four cardiovascular MR sequences, notably improving inversion-recovery T1 estimation.
- Phantom studies showed mean biases <30 ms for T1 and <1 ms for T2 between DeepFittingNet and curve-fitting.
- Excellent agreement (mean bias <6 ms) and comparable precision were observed in left ventricle and septum T1/T2 values in vivo.
Conclusions:
- DeepFittingNet effectively performs T1/T2 estimation for commonly used cardiovascular MR mapping sequences (MOLLI, SASHA, T2-prep bSSFP).
- The deep learning approach offers improved robustness for T1 estimation compared to traditional curve-fitting algorithms.
- DeepFittingNet presents a promising, accurate, and precise alternative for quantitative CMR analysis.
