## Fast, iterative image reconstruction for MRI in the presence of field inhomogeneities (2003)

### Cached

### Download Links

- [spin.ecn.purdue.edu]
- [www.eecs.umich.edu]
- DBLP

### Other Repositories/Bibliography

Venue: | IEEE TRANS. MED. IMAGING |

Citations: | 2 - 1 self |

### BibTeX

@ARTICLE{Sutton03fast,iterative,

author = {Bradley P. Sutton and Douglas C. Noll and Jeffrey A. Fessler},

title = {Fast, iterative image reconstruction for MRI in the presence of field inhomogeneities},

journal = {IEEE TRANS. MED. IMAGING},

year = {2003},

volume = {22},

number = {2},

pages = {178--188}

}

### OpenURL

### Abstract

In magnetic resonance imaging, magnetic field inhomogeneities cause distortions in images that are reconstructed by conventional fast Fourier trasform (FFT) methods. Several noniterative image reconstruction methods are used currently to compensate for field inhomogeneities, but these methods assume that the field map that characterizes the off-resonance frequencies is spatially smooth. Recently, iterative methods have been proposed that can circumvent this assumption and provide improved compensation for off-resonance effects. However, straightforward implementations of such iterative methods suffer from inconveniently long computation times. This paper describes a tool for accelerating iterative reconstruction of field-corrected MR images: a novel time-segmented approximation to the MR signal equation. We use a min–max formulation to derive the temporal interpolator. Speedups of around 60 were achieved by combining this temporal interpolator with a nonuniform fast Fourier transform with normalized root mean squared approximation errors of 0.07%. The proposed method provides fast, accurate, field-corrected image reconstruction even when the field map is not smooth.

### Citations

2073 |
Robust Statistics
- Huber
- 1981
(Show Context)
Citation Context ...t function used here is appropriate for Gaussian measurement noise. If non-Gaussian error “spikes” are present, then one could use a nonquadratic cost function to provide robustness to those outliers =-=[31]-=-, at the expense of increased computation. Alternatively, one could use other methods to detect those spikes, e.g., [32], then exclude the (7) update image update residual The dominant computation in ... |

151 |
Fast Fourier transforms for nonequispaced data
- Dutt, Rokhlin
- 1993
(Show Context)
Citation Context ...omputation time, with reported values of computation time/iteration ranging up to eight minutes [14]. Recently, accurate and fast nonuniform fast Fourier transform (NUFFT) methods have been developed =-=[17]-=-–[19] and these methods have been applied to MRI data with spiral k-space trajectories [20], [21]. The MR reconstruction problem is closely related to the problem of reconstructing a band-limited sign... |

102 |
Spatial resolution properties of penalized-likelihood image reconstruction methods: Spatial-invariant tomographs
- Fessler, Rogers
- 1996
(Show Context)
Citation Context ...an decrease the condition number of the image reconstruction problem and, therefore, speed convergence. We choose the parameter by examining the point spread function (PSF) of the reconstructed image =-=[30]-=-, preferably by choosing small enough to not significantly degrade the spatial resolution relative to the natural resolution associated with the k-space trajectory. The least-squares cost function use... |

100 |
On the fast Fourier transform of functions with singularities
- Beylkin
- 1995
(Show Context)
Citation Context ...for using trigonometric polynomials (complex exponentials) for finite-dimensional approximations in such problems, and proposed to use an iterative CG reconstruction method with the NUFFT approach of =-=[22]-=- at its core [23], [24]. In the MR context, this is essentially equivalent to the finite basis expansion we use in (3). In [25], an NUFFT-like algorithm, referred to as “reverse gridding,” was applied... |

96 | Selection of a convolution function for Fourier inversion using gridding [computerised tomography application
- JACKSON, MEYER, et al.
- 1991
(Show Context)
Citation Context ...truction is to interpolate the ’s onto a cartesian grid in spatial frequency space, after applying sample density compensation, and to then use an inverse FFT and deapodization to estimate samples of =-=[27]-=-. This gridding method, when combined with time segmentation of the field inhomogeneity effects, is a fast conjugate phase approach [3]. The combination of (1) and (2) form a continuous-to-discrete (C... |

90 | Nonuniform fast Fourier transforms using min-max interpolation
- Fessler, Sutton
(Show Context)
Citation Context ..., weighted by the coefficients . We can perform these inner FTs quickly and accurately using an NUFFT [19]. Our goal here is to choose the to minimize the error of approximation (9). In the spirit of =-=[18]-=- and [19], we propose to adopt a min–max criterion to optimize the temporal interpolation coefficients, for , i.e., for every point in the k-space readout. For any time , we choose the coefficients us... |

88 | SENSE: sensitivity encoding for fast MRI
- Pruessmann, Weiger, et al.
- 1999
(Show Context)
Citation Context ...he adjoint of the forward projector. Interest in iterative reconstruction methods has increased recently due to its utility in multiple coil noncartesian k-space sensitivity encoding (SENSE) problems =-=[15]-=-. Due to the complex aliasing pattern associated with undersampling k-space trajectories such as spirals, iterative methods that include coil sensitivity patterns in the projectors are necessary to re... |

57 | Conjugate-gradient preconditioning methods for shift-variant PET image reconstruction
- Fessler, Booth
- 1999
(Show Context)
Citation Context ...f higher off-resonance effects. Our results to date with circulant preconditioners have shown mild improvements in convergence rate. Preconditioners have also been designed for shift-variant problems =-=[42]-=- and such methods will be investigated for MRI in our future work. III. METHODS Three sets of studies were performed to evaluate the accuracy and utility of our min–max interpolated iterative reconstr... |

41 |
Computationally attractive reconstruction of bandlimited images from irregular samples
- Strohmer
- 1997
(Show Context)
Citation Context ...metric polynomials (complex exponentials) for finite-dimensional approximations in such problems, and proposed to use an iterative CG reconstruction method with the NUFFT approach of [22] at its core =-=[23]-=-, [24]. In the MR context, this is essentially equivalent to the finite basis expansion we use in (3). In [25], an NUFFT-like algorithm, referred to as “reverse gridding,” was applied in combination w... |

38 | Sampling density compensation in MRI: rationale and an iterative numerical solution. Magn Reson Med - JG, Menon - 1999 |

37 |
Preconditioning methods for improved convergence rates in iterative reconstruction
- Clinthorne, Pan, et al.
(Show Context)
Citation Context ...e algorithms can be accelerated by the use of an appropriate preconditioner, e.g., in (8). Circulant preconditioners have been shown to be effective in shift-invariant problems in tomographic imaging =-=[41]-=-. These preconditioners attempt to undo the blurring induced by applying the forward projector and its adjoint. A circulant preconditioner should be particularly helpful for MR reconstruction with sma... |

35 |
Advances in sensitivity encoding with arbitrary k-space trajectories
- Pruessmann, Weiger, et al.
- 2001
(Show Context)
Citation Context ... to use an iterative CG reconstruction method with the NUFFT approach of [22] at its core [23], [24]. In the MR context, this is essentially equivalent to the finite basis expansion we use in (3). In =-=[25]-=-, an NUFFT-like algorithm, referred to as “reverse gridding,” was applied in combination with the CG algorithm to speed up SENSE image reconstructions. These NUFFT methods have reduced the computation... |

30 |
A homogeneity correction method for magnetic resonance imaging with time-varying gradients
- Noll, Meyer, et al.
- 1991
(Show Context)
Citation Context ...ions [1], whereas spatial blur is induced in noncartesian k-space MR imaging (MRI) (using spirals, etc.) [2]. Many image reconstruction methods have been proposed to correct for the field distortions =-=[3]-=-–[7]. We focus on algorithms appropriate for conventional computers; optical implementations may also be feasible [8]. There are two components to most methods for field-corrected MR image reconstruct... |

27 | On the optimality of the gridding reconstruction algorithm
- Sedarat, Nishimura
(Show Context)
Citation Context ...struction and that the weights be the coefficients of the sampling density compensation function [25]. However, there has been some discussion on how to calculate optimal density compensation factors =-=[35]-=-–[39] and the iterative image reconstruction algorithm does not require and is not dependent upon these weights if an unweighted (i.e., weighting is unity) least squares approach is used instead. Also... |

26 |
Principles of Magnetic Resonance Imaging
- Nishimura
- 1996
(Show Context)
Citation Context ...econstruction. Simulation and human data experiments are described in Section III with the results given in Section IV. II. THEORY In MRI, ignoring relaxation effects, the signal equation is given by =-=[26]-=- where is the complex baseband signal at time during the readout, is the echo time, is a continuous function of the object’s transverse magnetization at location immediately following the spin prepara... |

22 |
Reconstruction of mr images from data acquired on a general nonregular grid by pseudoinverse calculation
- Walle, Barrett, et al.
(Show Context)
Citation Context ... The combination of (1) and (2) form a continuous-to-discrete (CD) mapping. This is clearly an ill-posed problem since there is an infinite collection of solutions, , that exactly match the data . In =-=[28]-=-, the pseudoinverse of this CD mapping was investigated for minimum-norm least-squares image reconstruction without field-correction. Although their approach was computationally intensive, the pseudoi... |

21 |
Off-resonance correction of MR images
- Schomberg
- 1999
(Show Context)
Citation Context ...d in this paper could be one component of an overall joint estimation procedure [9]. After a field map is obtained, one method of field-corrected image reconstruction, the conjugate phase method [3], =-=[6]-=-, [7], seeks to compensate for the phase accrual at each time point due to the off-resonance. This method, like most noniterative methods, relies on the assumption of a smooth field map. Timesegmented... |

18 |
Macovski A. Multifrequency interpolation for fast off-resonance correction. Magn Reson Med
- Man, JM
- 1997
(Show Context)
Citation Context ... [1], whereas spatial blur is induced in noncartesian k-space MR imaging (MRI) (using spirals, etc.) [2]. Many image reconstruction methods have been proposed to correct for the field distortions [3]–=-=[7]-=-. We focus on algorithms appropriate for conventional computers; optical implementations may also be feasible [8]. There are two components to most methods for field-corrected MR image reconstruction.... |

17 |
Rapid in vivo proton shimming
- Schneider, Glover
- 1991
(Show Context)
Citation Context ...the Institutional Review Board of the University of Michigan. For the human data, the field inhomogeneity map must be measured by acquiring two gradient echo images with slightly different echo times =-=[43]-=-. To minimize field inhomogeneity distortions in the images used to estimate the field map, we acquired a pair of 4-shot gradient echo images with ’s of 5 and 7 ms. This fieldmap was used to reconstru... |

15 | Deblurring for non-2D Fourier transform magnetic resonance imaging. Magnetic Resonance in Medicine - Noll, Pauly, et al. - 1992 |

12 |
Algebraic reconstruction for magnetic resonance imaging under B0 inhomogeneity
- Kadah, Hu
- 1998
(Show Context)
Citation Context ...r than using an inappropriately weighted CG algorithm, we use the conjugate phase image (reconstructed via a fast, density-compensated, time-segmented approach) as the initial estimate, . As noted in =-=[40]-=-, initializing with a good density-compensated conjugate phase image ensures that subsequent iterations will improve on this initial guess. Fig. 1. (a) (b) (a) Simulation object and (b) field map in H... |

11 | An inverse problem approach to the correction of distortion in EPI images
- Munger, Crelier, et al.
- 2000
(Show Context)
Citation Context ...t regularity of a time map required for iterative methods. Model-based iterative reconstruction methods have the potential to account for field maps that violate smoothness assumptions. Munger et al. =-=[12]-=- reported that iterative conjugategradient methods based on Fourier reconstructed echo-planar images outperform the conjugate-phase approach. Their sparsified system model is specific to cartesian tra... |

11 |
A Macovski. “Inhomogeneity correction using an estimated linear field map.” Magnetic resonance in medicine
- Irarrazabal, Meyer, et al.
- 1996
(Show Context)
Citation Context ...IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 22, NO. 2, FEBRUARY 2003 the frequency-segmentation method for reducing computation in the conjugate-phase approach for field inhomogeneity correction [33], =-=[34]-=-. The expression for this interpolator bears a striking resemblance to the “multifrequency interpolator” proposed by Man et al. [7]. However, the use of the two interpolators is quite different. The m... |

10 |
Tweig, “Iterative reconstruction of single-shot spiral mri with off resonance
- Harshbarger, B
- 1999
(Show Context)
Citation Context ... © 2003 IEEESUTTON et al.: FAST, ITERATIVE IMAGE RECONSTRUCTION FOR MRI IN THE PRESENCE OF FIELD INHOMOGENEITIES 179 rapidly varying inhomogeneity. The iterative reconstruction algorithm proposed in =-=[14]-=- was shown to provide significant improvements in image quality over noniterative methods even for field maps with discontinuities. Their method also can be used in an extended form to estimate more a... |

9 | Amin-max approach to the nonuniform N-dimensional FFT for rapid iterative reconstruction of MR images
- Sutton, Fessler, et al.
- 2001
(Show Context)
Citation Context ...14]. Recently, accurate and fast nonuniform fast Fourier transform (NUFFT) methods have been developed [17]–[19] and these methods have been applied to MRI data with spiral k-space trajectories [20], =-=[21]-=-. The MR reconstruction problem is closely related to the problem of reconstructing a band-limited signal from nonuniform samples. Strohmer argued compellingly for using trigonometric polynomials (com... |

8 |
Simulated phase evolution rewinding (SPHERE): A technique for reducing B inhomogeneity effects
- Kadah, Hu
- 1997
(Show Context)
Citation Context ... provides a rigorous analysis of the family of conjugate-phase methods for off-resonance correction of MR images, and concludes that segmented conjugate-phase methods are preferable to SPHERE methods =-=[5]-=-, at least for spiral imaging. Therefore, in this paper we focus on comparing our proposed iterative methods to the conjugate-phase method as the de facto standard for noniterative off-resonance corre... |

8 | Reconstructing MR images from undersampled data: data-weighing considerations - Pipe - 2000 |

7 |
Image restoration from nonuniform magnetic field influence for direct Fourier NMR imaging,” Phys
- Sekihara, Kuroda, et al.
- 1984
(Show Context)
Citation Context ...ain, cause image distortions in magnetic resonance (MR) images formed by conventional reconstruction methods. In spin-warp imaging, off-resonance effects cause spatial shifts and intensity variations =-=[1]-=-, whereas spatial blur is induced in noncartesian k-space MR imaging (MRI) (using spirals, etc.) [2]. Many image reconstruction methods have been proposed to correct for the field distortions [3]–[7].... |

6 | Reconstruction of MR images from data acquired on an arbitrary k-space trajectory using the same-image weight,” Magn - Qian, Lin, et al. - 2002 |

4 |
Correcting severe local inhomogeneity blurs by spatially variant deconvolution
- Man, Pauly, et al.
- 1995
(Show Context)
Citation Context ...pproach. Their sparsified system model is specific to cartesian trajectories like echo-planar, whereas the conjugate gradient (CG) approach considered here is applicable to any trajectory. Man et al. =-=[13]-=- described an iterative algorithm to remove the residual blur left over after conjugate phase reconstruction in regions with 0278-0062/03$17.00 © 2003 IEEESUTTON et al.: FAST, ITERATIVE IMAGE RECONST... |

4 |
Reconstruction techniques for magnetic resonance imaging
- Noll
- 1991
(Show Context)
Citation Context ...o182 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 22, NO. 2, FEBRUARY 2003 the frequency-segmentation method for reducing computation in the conjugate-phase approach for field inhomogeneity correction =-=[33]-=-, [34]. The expression for this interpolator bears a striking resemblance to the “multifrequency interpolator” proposed by Man et al. [7]. However, the use of the two interpolators is quite different.... |

3 |
Spiral sampling in magnetic resonance imaging—the effect of inhomogeneities
- Yudilevich, Stark
(Show Context)
Citation Context ... methods. In spin-warp imaging, off-resonance effects cause spatial shifts and intensity variations [1], whereas spatial blur is induced in noncartesian k-space MR imaging (MRI) (using spirals, etc.) =-=[2]-=-. Many image reconstruction methods have been proposed to correct for the field distortions [3]–[7]. We focus on algorithms appropriate for conventional computers; optical implementations may also be ... |

3 |
Rapid MR image acquisition in the presence of background gradients
- Noll
(Show Context)
Citation Context ...t smooth may be due to incorrect density compensation coefficients. Spatially varying density compensation may be necessary in those cases, restricting the application of methods to speed computation =-=[10]-=-. Iterative reconstruction methods do not require density compensation coefficients and are immune to discussions on how to calculate accurate density coefficients. Schomberg [6] provides a rigorous a... |

3 |
MacFall, “Correction of MR k-space data corrupted by spike noise
- Kao, R
- 2000
(Show Context)
Citation Context ... could use a nonquadratic cost function to provide robustness to those outliers [31], at the expense of increased computation. Alternatively, one could use other methods to detect those spikes, e.g., =-=[32]-=-, then exclude the (7) update image update residual The dominant computation in each iteration of the CG algorithm is computing and , where the superscript denotes complex conjugate transpose. Computi... |

3 |
Notes on direct and gridding-based Fourier inversion methods
- Schomberg
(Show Context)
Citation Context ...tion and that the weights be the coefficients of the sampling density compensation function [25]. However, there has been some discussion on how to calculate optimal density compensation factors [35]–=-=[39]-=- and the iterative image reconstruction algorithm does not require and is not dependent upon these weights if an unweighted (i.e., weighting is unity) least squares approach is used instead. Also, ass... |

3 |
Field-corrected imaging using joint estimation of image and field map
- Sutton, Fessler, et al.
(Show Context)
Citation Context ...ns were about 4 s each. (a) Uncorrected. (b) Conjugate phase. (c) Fast iterative. (d) Field map (Hz). simultaneously estimate an undistorted field map and provide a better field-corrected image [14], =-=[44]-=-. V. DISCUSSION We have presented a method that allows fast, iterative reconstruction of field-corrected MR images. By combining the NUFFT with time segmentation using a min–max temporal interpolator,... |

3 |
Echo-volume imaging
- Song, Wong, et al.
- 1994
(Show Context)
Citation Context ...The ability to accurately compensate for off-resonance effects as demonstrated here may increase the feasibility of using other acquisition methods with long readout times, such as echovolume imaging =-=[46]-=-. Although this paper has focused on MR image reconstruction in the presence of field inhomogeneities, the general approach is also applicable to image reconstruction with compensation for other sourc... |

2 | Simultaneous estimation of image and inhomogeneity field map
- Sutton, Noll, et al.
(Show Context)
Citation Context ...or desirable to couple the field-map estimation and image reconstruction procedures. In such cases, the methods described in this paper could be one component of an overall joint estimation procedure =-=[9]-=-. After a field map is obtained, one method of field-corrected image reconstruction, the conjugate phase method [3], [6], [7], seeks to compensate for the phase accrual at each time point due to the o... |

2 |
Fast image reconstruction for sensitivity encoded spiral imaging
- Kannengießer, Pauly, et al.
(Show Context)
Citation Context ...utes [14]. Recently, accurate and fast nonuniform fast Fourier transform (NUFFT) methods have been developed [17]–[19] and these methods have been applied to MRI data with spiral k-space trajectories =-=[20]-=-, [21]. The MR reconstruction problem is closely related to the problem of reconstructing a band-limited signal from nonuniform samples. Strohmer argued compellingly for using trigonometric polynomial... |

2 |
Model-based maximum likelihood estimation for phase- and frequency-encoded magnetic-resonance-imaging data
- Miller, Schaewe, et al.
- 1995
(Show Context)
Citation Context ...l model (1). However, many aspects of the algorithms we have described are also applicable to problems where both spin density and spin relaxation are estimated from multiecho measurements [9], [16], =-=[49]-=-, [50]. The framework for the min–max time interpolation provided by (16) can be extended to include relaxation effects, such as . The simplifications that resulted in (18) are not available in this c... |

1 |
Optical implementation of spiral EPI field homogeneity correction and the decoupled automated rotation and translation algorithm (DFT-DART
- NessAiver
(Show Context)
Citation Context ...age reconstruction methods have been proposed to correct for the field distortions [3]–[7]. We focus on algorithms appropriate for conventional computers; optical implementations may also be feasible =-=[8]-=-. There are two components to most methods for field-corrected MR image reconstruction. The first procedure is to obtain an estimate of the field map that quantifies the spatial distribution of magnet... |

1 |
Rosette trajectories for spectrally selective MR imaging
- “Multishot
- 1997
(Show Context)
Citation Context ...k-space point to a unique acquisition time. Our proposed iterative methods do not require any such assumption and are, therefore, applicable to self-intersecting k-space trajectories such as rosettes =-=[11]-=-. Nor are any assumptions about regularity of a time map required for iterative methods. Model-based iterative reconstruction methods have the potential to account for field maps that violate smoothne... |

1 |
Simultaneous estimation of I , R , and field map using a multi-echo spiral acquisition
- Peltier, Sutton, et al.
- 2002
(Show Context)
Citation Context ...on field inhomogeneities, one can also apply iterative image reconstruction methods to compensate for other physical phenomena such as deviations in k-space trajectory and relaxation effects, such as =-=[16]-=-. The principal drawback of iterative reconstruction methods has been computation time, with reported values of computation time/iteration ranging up to eight minutes [14]. Recently, accurate and fast... |

1 |
min–max approach to the multidimensional nonuniform FFT: Application to tomographic image reconstruction
- “A
(Show Context)
Citation Context ...ation time, with reported values of computation time/iteration ranging up to eight minutes [14]. Recently, accurate and fast nonuniform fast Fourier transform (NUFFT) methods have been developed [17]–=-=[19]-=- and these methods have been applied to MRI data with spiral k-space trajectories [20], [21]. The MR reconstruction problem is closely related to the problem of reconstructing a band-limited signal fr... |

1 |
analysis of the nonuniform sampling problem
- “Numerical
- 2000
(Show Context)
Citation Context ... polynomials (complex exponentials) for finite-dimensional approximations in such problems, and proposed to use an iterative CG reconstruction method with the NUFFT approach of [22] at its core [23], =-=[24]-=-. In the MR context, this is essentially equivalent to the finite basis expansion we use in (3). In [25], an NUFFT-like algorithm, referred to as “reverse gridding,” was applied in combination with th... |

1 |
Concomitant gradient field effects
- King, Ganin, et al.
- 1999
(Show Context)
Citation Context ...neral approach is also applicable to image reconstruction with compensation for other sources of undesired (but known) spin phase accrual, such as eddy currents and concomitant gradient effects [47], =-=[48]-=-. An iterative method based on an explicit signal model like (1) should yield more accurate images compared with conventional approaches to compensating for such effects. We have ignored spin-spin rel... |