Compressed sensing MRI with singular value decomposition-based sparsity basis

Hong, Mingjian; Yu, Yeyang; Wang, Hua; F, Liu; Crozier, Stuart

doi:10.1088/0031-9155/56/19/010

Cited by 59 publications

(46 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Sparse representation can be explored in a specific transform domain or generally in a dictionary-based subspace [16]. Classic fast CS-MRI uses predefined and fixed sparsifying transforms, e.g., total variation (TV) [17], [18], [19], discrete cosine transforms [20], [21], [22] and discrete wavelet transforms [23], [24], [25]. In addition, this has been extended to a more flexible sparse representation learnt directly from data using dictionary learning [26], [27], [28].…”

Section: Related Work and Our Contributions A Classic Model-basementioning

confidence: 99%

DAGAN: Deep De-Aliasing Generative Adversarial Networks for Fast Compressed Sensing MRI Reconstruction

Yang

Dong

et al. 2018

IEEE Trans. Med. Imaging

886

628

View full text Add to dashboard Cite

Abstract-Compressed Sensing Magnetic Resonance Imaging (CS-MRI) enables fast acquisition, which is highly desirable for numerous clinical applications. This can not only reduce the scanning cost and ease patient burden, but also potentially reduce motion artefacts and the effect of contrast washout, thus yielding better image quality. Different from parallel imaging based fast MRI, which utilises multiple coils to simultaneously receive MR signals, CS-MRI breaks the Nyquist-Shannon sampling barrier to reconstruct MRI images with much less required raw data. This paper provides a deep learning based strategy for reconstruction of CS-MRI, and bridges a substantial gap between conventional non-learning methods working only on data from a single image, and prior knowledge from large training datasets. In particular, a novel conditional Generative Adversarial Networks-based model (DAGAN) is proposed to reconstruct CS-MRI. In our DAGAN architecture, we have designed a refinement learning method to stabilise our U-Net based generator, which provides an endto-end network to reduce aliasing artefacts. To better preserve texture and edges in the reconstruction, we have coupled the adversarial loss with an innovative content loss. In addition, we incorporate frequency domain information to enforce similarity in both the image and frequency domains. We have performed comprehensive comparison studies with both conventional CS-MRI reconstruction methods and newly investigated deep learning approaches. Compared to these methods, our DAGAN method provides superior reconstruction with preserved perceptual image details. Furthermore, each image is reconstructed in about 5 ms, which is suitable for real-time processing.

show abstract

Section: Related Work and Our Contributions A Classic Model-basementioning

confidence: 99%

DAGAN: Deep De-Aliasing Generative Adversarial Networks for Fast Compressed Sensing MRI Reconstruction

Yang

Dong

et al. 2018

IEEE Trans. Med. Imaging

886

628

View full text Add to dashboard Cite

show abstract

“…Recently, a range of sparsity bases in spatial and temporal dimensions were proposed to implement the sparsifying transform, such as discrete Wavelet transform [14], discrete cosine transform [15], total variation [16], one-dimensional Fourier transform [17], KLT/PCA transform [18], singular value decomposition [19], motion estimation [20] and dictionary learning [21]. Compressed sensing (CS) method has been successfully applied in static and dynamic MRI studies.…”

Section: Introductionmentioning

confidence: 99%

Improved l1-SPIRiT using 3D walsh transform-based sparsity basis

Zhao¹,

F²,

Jiang³

et al. 2014

Magnetic Resonance Imaging

Self Cite

View full text Add to dashboard Cite

l1-SPIRiT is a fast magnetic resonance imaging (MRI) method which combines parallel imaging (PI)with compressed sensing (CS) by performing a joint l1-norm and l2-norm optimization procedure.The original l1-SPIRiT method uses two-dimensional (2D) Wavelet transform to exploit the intra-coil data redundancies and a joint sparsity model to exploit the inter-coil data redundancies. In this work, we propose to stack all the coil images into a three-dimensional (3D) matrix, and then a novel 3D Walsh transform-based sparsity basis is applied to simultaneously reduce the intra-coil and inter-coil data redundancies. Both the 2D Wavelet transform-based and the proposed 3D Walsh transform-based sparsity bases were investigated in the l1-SPIRiT method. The experimental results show that the proposed 3D Walsh transform-based l1-SPIRiT method outperformed the original l1-SPIRiT in terms of image quality and computational efficiency.

show abstract

“…Irregular sampling patterns, such as variable density k-space sampling, are effective ways to improve incoherence between the sparse transform and sampling domains making aliasing interference less prominent. The second approach attempts to use a range of sparsity bases in spatial and temporal dimensions, such as discrete wavelet transform, curvelet transform [98] and singular value decomposition based transforms [99] to provide sufficient sparsity for faithful reconstruction using a subset of the largest transform coefficient. The third category concentrates on the nonlinear optimization methods for signal recovery [8,10,100].…”

Section: Current Challenges In Cs Mrimentioning

confidence: 99%

“…Recently, the CS theory has been considerably improved and can be used to further promote the reconstructed image quality of CS MRI. For example, the novel sparsity bases in CS [79,80,92,99] have the potential in to provide optimal sparse representation of objects to retain more details with the same number of samples. In the static CS MRI, shearlet and curvelet [146,147] sparsity transforms can preserve the edges and features in image reconstruction because of the excellent directional selectivity and localization property.…”

Section: Improvement To Current Solutionsmentioning

confidence: 99%

“…For example, the discrete wavelet transform [8], curvelet transform [98], walsh transform [109] and singular value decomposition-based transforms [80,99] are commonly used as sparsity transform. However, without sufficient k-space data, sparse coefficients transformed from the sampling signals cannot faithfully represent the original object and thus the quality of the reconstructed images is degraded.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Improving the image quality in compressed sensing MRI by the exploitation of data properties

Yang¹

View full text Add to dashboard Cite

Minimizing scan time without compromising image quality has been the main thrust of magnetic resonance imaging (MRI) since the establishment of the field. Tremendous effort has been put into MRI systems in pursuit of faster MR imaging techniques, which can substantially facilitate clinical diagnoses and improve the patient experience. Traditional MRI accelerated approaches mainly focused on hardware improvements to improve the speed of scanning. These methods are still governed by Nyquist-Shannon sampling rate. Hence, when the acceleration rate is pushed beyond the theoretical limit, aliasing artefacts are introduced which degrade the reconstructed image quality and are difficult to eliminate via current methods. Recently, it has been found that a newly developed signal processing theory, compressed sensing (CS), can be used to create high-resolution signal data sets from sparse samples, despite violating the classical Nyquist-Shannon sampling criteria. The application of CS to MRI opens up an appealing avenue to offer potentially significant scan time reductions. CS reconstructs MR images from incomplete k-space measurements. Taking advantage of the fact that MR images have sparse representations under certain mathematical transformations, CS overcomes the distortions resulting from the under-sampled k-space data. Significant progress has been made in the development of CS MRI. However, there are two major remaining challenges to its full adoption in clinical applications: (1) the small amount of under-sampled data (for example, less than 1/8 of the whole k-space data) will inherently introduce artefacts in the reconstructed image and compromise diagnosis accuracy for the reconstructed images; (2) owing to hardware limits, it is difficult to realise two-dimension random under-sampling, which is favoured by CS operation.Practically, the random phase-encode under-sampling pattern has been widely implemented.However, it introduces coherent aliasing artefacts that are hard to eliminate using the CS theory.In this thesis, several techniques are proposed that can improve the quality of images reconstructed by CS MRI. These techniques have the potential to facilitate faithful reconstruction of CS MRI in clinical applications. A novel two-stage reconstruction scheme is introduced in Chapter 3. In this method, the under-sampled k-space data is segmented into low-frequency and high-frequency parts.Then, in stage one, using dense measurements, the low-frequency region of k-space data is faithfully reconstructed. The fully reconstituted low-frequency k-space data from the first stage is then combined with the under-sampled high-frequency k-space data to complete the second stage reconstruction of the whole image. With this two-stage approach, each reconstruction inherently incorporates a lower data under-sampling rate than conventional approaches. Because the restricted isometric property is easier to satisfy than conventional CS method, the reconstruction consequently ii produces lower residual errors in each step...

show abstract

Compressed sensing MRI with singular value decomposition-based sparsity basis

Cited by 59 publications

References 16 publications

DAGAN: Deep De-Aliasing Generative Adversarial Networks for Fast Compressed Sensing MRI Reconstruction

DAGAN: Deep De-Aliasing Generative Adversarial Networks for Fast Compressed Sensing MRI Reconstruction

Improved l1-SPIRiT using 3D walsh transform-based sparsity basis

Improving the image quality in compressed sensing MRI by the exploitation of data properties

Contact Info

Product

Resources

About