Multi-Contrast Super-Resolution MRI Through a Progressive Network

Lyu, Qing; Shan, Hongming; Steber, C.; Helis, Corbin A.; Whitlow, Chris; Chan, Michael D.; Wang, Ge

doi:10.1109/tmi.2020.2974858

Cited by 114 publications

(64 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In medical imaging, many deep learning-based frameworks have been introduced for feature extraction, anatomical landmark detection, and segmentation [15] – [17] . Recently, deep learning-based single image SR methods for medical imaging have been actively explored [18] – [28] .…”

Section: Introductionmentioning

confidence: 99%

Autoencoder-Inspired Convolutional Network-Based Super-Resolution Method in MRI

Park

Gach

Kim

et al. 2021

IEEE J. Transl. Eng. Health Med.

View full text Add to dashboard Cite

Objective: To introduce an MRI in-plane resolution enhancement method that estimates High-Resolution (HR) MRIs from Low-Resolution (LR) MRIs. Method & Materials: Previous CNN-based MRI super-resolution methods cause loss of input image information due to the pooling layer. An Autoencoder-inspired Convolutional Network-based Super-resolution (ACNS) method was developed with the deconvolution layer that extrapolates the missing spatial information by the convolutional neural network-based nonlinear mapping between LR and HR features of MRI. Simulation experiments were conducted with virtual phantom images and thoracic MRIs from four volunteers. The Peak Signal-to-Noise Ratio (PSNR), Structure SIMilarity index (SSIM), Information Fidelity Criterion (IFC), and computational time were compared among: ACNS; Super-Resolution Convolutional Neural Network (SRCNN); Fast Super-Resolution Convolutional Neural Network (FSRCNN); Deeply-Recursive Convolutional Network (DRCN). Results: ACNS achieved comparable PSNR, SSIM, and IFC results to SRCNN, FSRCNN, and DRCN. However, the average computation speed of ACNS was 6, 4, and 35 times faster than SRCNN, FSRCNN, and DRCN, respectively under the computer setup used with the actual average computation time of 0.15 s per \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$100\times100$ \end{document}

show abstract

Section: Introductionmentioning

confidence: 99%

Autoencoder-Inspired Convolutional Network-Based Super-Resolution Method in MRI

Park

Gach

Kim

et al. 2021

IEEE J. Transl. Eng. Health Med.

View full text Add to dashboard Cite

show abstract

“…More recently, deep neural networks have been exploited for fast MC images synthesis 25–27 and reconstruction 28–37 . Rather than using fixed, handcrafted extractors, the networks learn the parameters of convolutional kernels for the shareable feature representations.…”

Section: Introductionmentioning

confidence: 99%

“…The features are extracted by the shallow layer, which represents the sharing structure of the original images. On the other hand, networks in 35–37 attempt to extract high‐level semantic features as shareable information using stacked convolutional operations for different contrast images before concatenating them in a deep layer. Promising reconstruction results have been reported in these works, demonstrating that the network can explore MC features with the learnable kernels.…”

Section: Introductionmentioning

confidence: 99%

“…The feature sharing is the core in MC image reconstruction; it can, however, be further improved beyond the operations implemented in the existing networks. First, the current MC features are expressed in a single resolution extracted from a single layer in existing algorithms 28–37 . For example, Lyu et al 36 assumed and proved that the high‐level, semantic MC features can lead to better reconstruction results compared with the low‐level MC features.…”

Section: Introductionmentioning

confidence: 99%

“…First, the current MC features are expressed in a single resolution extracted from a single layer in existing algorithms 28–37 . For example, Lyu et al 36 assumed and proved that the high‐level, semantic MC features can lead to better reconstruction results compared with the low‐level MC features. In fact, the local features, captured by the shallow layers, and high‐level abstract information, captured by the deep layers, represent complementary shareable structures, 38–40 which need to be exploited jointly for better results.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Optimizing multicontrast MRI reconstruction with shareable feature aggregation and selection

et al. 2021

View full text Add to dashboard Cite

This paper proposes a new method for optimizing feature sharing in deep neural network‐based, rapid, multicontrast magnetic resonance imaging (MC‐MRI). Using the shareable information of MC images for accelerated MC‐MRI reconstruction, current algorithms stack the MC images or features without optimizing the sharing protocols, leading to suboptimal reconstruction results. In this paper, we propose a novel feature aggregation and selection scheme in a deep neural network to better leverage the MC features and improve the reconstruction results. First, we propose to extract and use the shareable information by mapping the MC images into multiresolution feature maps with multilevel layers of the neural network. In this way, the extracted features capture complementary image properties, including local patterns from the shallow layers and semantic information from the deep layers. Then, an explicit selection module is designed to compile the extracted features optimally. That is, larger weights are learned to incorporate the constructive, shareable features; and smaller weights are assigned to the unshareable information. We conduct comparative studies on publicly available T2‐weighted and T2‐weighted fluid attenuated inversion recovery brain images, and the results show that the proposed network consistently outperforms existing algorithms. In addition, the proposed method can recover the images with high fidelity under 16 times acceleration. The ablation studies are conducted to evaluate the effectiveness of the proposed feature aggregation and selection mechanism. The results and the visualization of the weighted features show that the proposed method does effectively improve the usage of the useful features and suppress useless information, leading to overall enhanced reconstruction results. Additionally, the selection module can zero‐out repeated and redundant features and improve network efficiency.

show abstract

3D‐SRDM: 3D super‐resolution of MRI volumes based on diffusion model

Wu,

Chen,

2024

Int J Imaging Syst Tech

View full text Add to dashboard Cite

Brain magnetic resonance imaging (MRI) is crucial for diagnosing and understanding neurological disorders, but inherent limitations hinder the visualization of fine details in brain structures. The emergence of super‐resolution techniques, especially deep‐learning methods, has improved imaging quality of MRI, by increasing MRI spatial resolution. At present, deep‐learning algorithms mostly performed super‐resolution on 2D MRI images. However, considering 3D nature of MRI, 3D models are more suitable for brain MRI super‐resolution. To achieve finer brain structural details, this study proposes a 3D brain MRI super‐resolution method based on diffusion model (3D‐SRDM), which is a fast and easily trainable neural network for the generation of high‐resolution brain MRI images. In our 3D‐SRDM model, the self‐attention module in U‐Net is replaced with 3D spatial attention mechanism. The network structure of 3D‐SRDM is optimized to reduce training parameters. Moreover, accelerated sampling from denoising diffusion implicit model is also incorporated to reduce time consumption. By these optimizations, compared with original diffusion model, the proposed model can achieve about 10‐ and 5‐fold speed increase at 4× and 8× super‐resolution of 3D brain MRI volumes, respectively, almost without affecting image quality. Thus, 3D‐SRDM has potential application value in efficiently generating high‐resolution 3D brain MRI images, thus facilitating the doctors' diagnosis.

show abstract

Multi-Contrast Super-Resolution MRI Through a Progressive Network

Cited by 114 publications

References 46 publications

Autoencoder-Inspired Convolutional Network-Based Super-Resolution Method in MRI

Autoencoder-Inspired Convolutional Network-Based Super-Resolution Method in MRI

Optimizing multicontrast MRI reconstruction with shareable feature aggregation and selection

3D‐SRDM: 3D super‐resolution of MRI volumes based on diffusion model

Contact Info

Product

Resources

About