2021
DOI: 10.1002/nbm.4527
|View full text |Cite
|
Sign up to set email alerts
|

Fast and accurate modeling of transient‐state, gradient‐spoiled sequences by recurrent neural networks

Abstract: Funding information China Scholarship Council Fast and accurate modeling of MR signal responses are typically required for various quantitative MRI applications, such as MR fingerprinting. This work uses a new extended phase graph (EPG)-Bloch model for accurate simulation of transient-state, gradient-spoiled MR sequences, and proposes a recurrent neural network (RNN) as a fast surrogate of the EPG-Bloch model for computing large-scale MR signals and derivatives. The computational efficiency of the RNN model is… Show more

Help me understand this report
View preprint versions

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
2
1

Citation Types

0
16
0

Year Published

2022
2022
2024
2024

Publication Types

Select...
5
2

Relationship

3
4

Authors

Journals

citations
Cited by 13 publications
(16 citation statements)
references
References 39 publications
0
16
0
Order By: Relevance
“…There is always a trade-off between the computation time and generalization capability when choosing surrogate models. We also explored surrogate models with much better generalization capability which could work for sequences with arbitrary RF flip-angle trains, however, the proposed RNN model in [31] is at least one order of magnitude slower than the model we used here. In the future, surrogate models can be trained to more complicated MR physics effects, for example water-fat effects [32] and magnetization transfer [33].…”
Section: Discussionmentioning
confidence: 99%
See 1 more Smart Citation
“…There is always a trade-off between the computation time and generalization capability when choosing surrogate models. We also explored surrogate models with much better generalization capability which could work for sequences with arbitrary RF flip-angle trains, however, the proposed RNN model in [31] is at least one order of magnitude slower than the model we used here. In the future, surrogate models can be trained to more complicated MR physics effects, for example water-fat effects [32] and magnetization transfer [33].…”
Section: Discussionmentioning
confidence: 99%
“…In the future, surrogate models can be trained to more complicated MR physics effects, for example water-fat effects [32] and magnetization transfer [33]. Surrogate models with good generalization capability could also be incorporated for working with different sequence parameters without the need for retraining [31].…”
Section: Discussionmentioning
confidence: 99%
“…• A 2D axial slice of the brain of two healthy volunteers (with approved consent according to the guidelines of the ethics committee). The quantitative 𝑇 1 and 𝑇 2 maps for each experimental setup were obtained applying the MR-STAT reconstruction as outlined in [15], [36], and [37].…”
Section: Analytical Tools (Noise Spectrum Diagrams) Derived Frommentioning
confidence: 99%
“…RNNs, (see Figure 2 ) are capable of memorizing temporal structures within sequences and are therefore good candidates for dictionary generation in MRF. For instance, Liu et al ( 63 ) propose a RNN as a surrogate model for dictionary generation for non-cardiac MRF. A novel feature of this network is that it is capable of modeling MRF signal evolutions resulting from different sequence parameters, in addition to encoding dependencies on tissue parameters.…”
Section: Artificial Intelligence In Cardiac Mrfmentioning
confidence: 99%
“…The RNN proposed by Liu et at. ( 63 ) for dictionary generation is also employed to develop a computationally efficient method to solve the CRLB optimization. In their work, they optimize a flip angle train of an MRF sequence given two target tissues by computing the 14,000 necessary magnetization signals and their derivatives with their proposed RNN in ~10s, a reduction of two orders of magnitude in runtime.…”
Section: Artificial Intelligence In Cardiac Mrfmentioning
confidence: 99%