One-Shot Generative Prior in Hankel-k-Space for Parallel Imaging Reconstruction

Peng, Hong; Jiang, Chen; Cheng, Jiang; Zhang, Minghui; Wang, Rongpin; Liu, Qiegen

doi:10.1109/tmi.2023.3288219

Cited by 11 publications

(5 citation statements)

References 47 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, generative model has drawn considerable attention in the field of MRI. [20][21][22][23][24][25] Leveraging distribution prior, generative models exhibit promise in enhancing acceleration while retaining intricate details. The data distribution can be learned implicitly by directly representing the sampling process like generative adversarial network (GAN) 26 or explicitly by representing the probability density/mass like Bayesian networks 27 and score-based generative model 28 that estimates the gradients of the data distribution.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Integrating data distribution prior via Langevin dynamics for end‐to‐end MR reconstruction

Cheng,

Cui,

Zhu

et al. 2024

Magnetic Resonance in Med

Self Cite

View full text Add to dashboard Cite

PurposeTo develop a novel deep learning‐based method inheriting the advantages of data distribution prior and end‐to‐end training for accelerating MRI.MethodsLangevin dynamics is used to formulate image reconstruction with data distribution before facilitate image reconstruction. The data distribution prior is learned implicitly through the end‐to‐end adversarial training to mitigate the hyper‐parameter selection and shorten the testing time compared to traditional probabilistic reconstruction. By seamlessly integrating the deep equilibrium model, the iteration of Langevin dynamics culminates in convergence to a fix‐point, ensuring the stability of the learned distribution.ResultsThe feasibility of the proposed method is evaluated on the brain and knee datasets. Retrospective results with uniform and random masks show that the proposed method demonstrates superior performance both quantitatively and qualitatively than the state‐of‐the‐art.ConclusionThe proposed method incorporating Langevin dynamics with end‐to‐end adversarial training facilitates efficient and robust reconstruction for MRI. Empirical evaluations conducted on brain and knee datasets compellingly demonstrate the superior performance of the proposed method in terms of artifact removing and detail preserving.

show abstract

Section: Introductionmentioning

confidence: 99%

“…Recently, generative model has drawn considerable attention in the field of MRI 20–25 . Leveraging distribution prior, generative models exhibit promise in enhancing acceleration while retaining intricate details.…”

Section: Introductionmentioning

confidence: 99%

Integrating data distribution prior via Langevin dynamics for end‐to‐end MR reconstruction

Cheng,

Cui,

Zhu

et al. 2024

Magnetic Resonance in Med

Self Cite

View full text Add to dashboard Cite

show abstract

“…Desai et al proposed the Noise2Recon for SNR-robust accelerated MRI reconstruction that can use both fully sampled (labeled) and undersampled (unlabeled) scans (Desai et al 2023). Peng et al proposed a novel Hankel-k-space generative model (HKGM) (Peng et al 2023) to learn to generate samples from only one case of fully sampled k-space data. Korkmaz et al introduced a zero-shot learned adversarial transformers (SLATER)-based unsupervised MRI reconstruction method without the need for paired training datasets, which still required fully-sampled MRI data to learn the MRI prior in a pre-training phase (Korkmaz et al 2022).…”

Section: Introductionmentioning

confidence: 99%

“…The reconstruction results of DFUSNN and HKGM are summarized in table 10. It should be noted that the reconstruction results of HKGM are obtained fromPeng et al (2023). From table 10, it is evident that DFUSNN outperforms HKGM in all cases, demonstrating better reconstruction performance.…”

mentioning

confidence: 95%

See 1 more Smart Citation

DFUSNN: zero-shot dual-domain fusion unsupervised neural network for parallel MRI reconstruction

Chen,

Duan,

Ren

et al. 2024

Phys. Med. Biol.

View full text Add to dashboard Cite

Objective. Recently, deep learning models have been used to reconstruct parallel magnetic resonance (MR) images from undersampled k-space data. However, most existing approaches depend on large databases of fully sampled MR data for training, which can be challenging or sometimes infeasible to acquire in certain scenarios. The goal is to develop an effective alternative for improved reconstruction quality that does not rely on external training datasets. Approach. We introduce a novel zero-shot dual-domain fusion unsupervised neural network (DFUSNN) for parallel MR imaging reconstruction without any external training datasets. We employ the Noise2Noise (N2N) network for the reconstruction in the k-space domain, integrates phase and coil sensitivity smoothness priors into the k-space N2N network, and uses an early stopping criterion to prevent overfitting. Additionally, we propose a dual-domain fusion method based on Bayesian optimization to enhance reconstruction quality efficiently. Results. Simulation experiments conducted on three datasets with different undersampling patterns showed that the DFUSNN outperforms all other competing unsupervised methods and the one-shot Hankel-k-space generative model (HKGM). The DFUSNN also achieves comparable results to the supervised Deep-SLR method. Significance. The novel DFUSNN model offers a viable solution for reconstructing high quality MR images without the need for external training datasets, thereby overcoming a major hurdle in scenarios where acquiring fully sampled MR data is difficult.

show abstract

Diffusion model based on generalized map for accelerated MRI

Xiao,

Lu,

et al. 2024

NMR in Biomedicine

View full text Add to dashboard Cite

In recent years, diffusion models have made significant progress in accelerating magnetic resonance imaging. Nevertheless, it still has inherent limitations, such as prolonged iteration times and sluggish convergence rates. In this work, we present a novel generalized map generation model based on mean‐reverting SDE, called GM‐SDE, to alleviate these shortcomings. Notably, the core idea of GM‐SDE is optimizing the initial values of the iterative algorithm. Specifically, the training process of GM‐SDE diffuses the original k‐space data to an intermediary degraded state with fixed Gaussian noise, while the reconstruction process generates the data by reversing this process. Based on the generalized map, three variants of GM‐SDE are proposed to learn k‐space data with different structural characteristics to improve the effectiveness of model training. GM‐SDE also exhibits flexibility, as it can be integrated with traditional constraints, thereby further enhancing its overall performance. Experimental results showed that the proposed method can reduce reconstruction time and deliver excellent image reconstruction capabilities compared to the complete diffusion‐based method.

show abstract

One-Shot Generative Prior in Hankel-k-Space for Parallel Imaging Reconstruction

Cited by 11 publications

References 47 publications

Integrating data distribution prior via Langevin dynamics for end‐to‐end MR reconstruction

Integrating data distribution prior via Langevin dynamics for end‐to‐end MR reconstruction

DFUSNN: zero-shot dual-domain fusion unsupervised neural network for parallel MRI reconstruction

Diffusion model based on generalized map for accelerated MRI

Contact Info

Product

Resources

About