## Toeplitz-based iterative image reconstruction for MRI with correction for magnetic field inhomogeneity (2005)

Venue: | IEEE Trans. Signal Process |

Citations: | 19 - 4 self |

### BibTeX

@ARTICLE{Fessler05toeplitz-basediterative,

author = {Jeffrey A. Fessler and Senior Member and Sangwoo Lee and Valur T. Olafsson and Hugo R. Shi and Student Member and Student Member and Student Member and Douglas C. Noll Member},

title = {Toeplitz-based iterative image reconstruction for MRI with correction for magnetic field inhomogeneity},

journal = {IEEE Trans. Signal Process},

year = {2005},

volume = {53},

pages = {3393--3402}

}

### Years of Citing Articles

### OpenURL

### Abstract

Abstract — In some types of magnetic resonance (MR) imaging, particularly functional brain scans, the conventional Fourier model for the measurements is inaccurate. Magnetic field inhomogeneities, caused by imperfect main fields and by magnetic susceptibility variations, induce distortions in images that are reconstructed by conventional Fourier methods. These artifacts hamper the use of functional MR imaging (fMRI) in brain regions near air/tissue interfaces. Recently, iterative methods that combine the conjugate gradient (CG) algorithm with nonuniform FFT (NUFFT) operations have been shown to provide considerably improved image quality relative to the conjugatephase method. However, for non-Cartesian k-space trajectories, each CG-NUFFT iteration requires numerous k-space interpolations, operations that are computationally expensive and poorly suited to fast hardware implementations. This paper proposes a faster iterative approach to field-corrected MR image reconstruction based on the CG algorithm and certain Toeplitz matrices. This CG-Toeplitz approach requires k-space interpolations only for the initial iteration; thereafter only FFTs are required. Simulation results show that the proposed CG-Toeplitz approach produces equivalent image quality as the CG-NUFFT method with significantly reduced computation time. Index Terms — fMRI imaging, spiral trajectory, magnetic susceptibility, non-Cartesian sampling I.

### Citations

200 |
Asymptotically optimal block quantization
- Gersho
- 1979
(Show Context)
Citation Context ...where they are most needed based on the rate map histogram. We achieve this by using the asymptotic theory of quantization, which specifies the optimal density of centroids for high-rate quantization =-=[50]-=-. C. Generalized approximations Both “time segmentation” and “frequency segmentation” lead to approximations of the form (14), and both enable the efficient implementation (15). Thus, from the point o... |

132 | Conjugate gradient methods for Toeplitz systems
- Chan, Ng
- 1996
(Show Context)
Citation Context ...o (13) as the “Toeplitz approach.” The primary bottleneck in using (13) is multiplication of T by x each iteration. If T were Toeplitz, then this could be done efficiently using wellknown FFT methods =-=[47]-=-, as has been proposed previously for iterative MR image reconstruction [40], [41]. Here, T is not Toeplitz due to the rate map z, so we will introduce approximations. The next section first examines ... |

85 | Nonuniform fast Fourier transforms using min-max interpolation
- Fessler, Sutton
- 2003
(Show Context)
Citation Context ... The computational bottleneck in (12) is calculating the matrix-vector products Ax and A ′ r, where r denotes the residual y − Ax. We previously used the above gradient expression and combined NUFFTs =-=[46]-=- with temporal interpolation based on a “time-segmentation” approximation [5] so as to be able to compute efficiently Ax and A ′ r [21]. We refer to (12) as the “NUFFT approach.” An alternative, mathe... |

83 | Correction for geometric distortion in echo planar images from B0 field variations. Magn Reson Med - Jezzard, Balaban - 1995 |

76 |
Functional brain mapping by blood oxygenation level-dependent contrast magnetic resonance imaging: A comparison of signal characteristics with a biophysical model
- Ogawa, Menon, et al.
- 1993
(Show Context)
Citation Context ...eads to geometric distortions in echo-planar imaging, and blurring and artifacts when imaging with non-Cartesian trajectories. These degradations can be severe in brain scans based on the BOLD effect =-=[2]-=-, hampering the use of fMRI in brain regions near air/tissue interfaces. Numerous solutions have been proposed based both on data acquisition strategies and reconstruction methods [3]–[22]. In the pre... |

52 | Conjugate-gradient preconditioning methods for shift-variant PET image reconstruction
- Fessler, Booth
- 1999
(Show Context)
Citation Context ...e “weighted sum” of Toeplitz matrices in (26). An open question for future work is how to precondition this sum effectively; preconditioners have been developed for other shift-variant problems [47], =-=[55]-=-. • The model (6) assumes that the rate map is constant over each voxel. To compensate for within-voxel field gradients, one can use smaller voxels [44]. This increases computation, so an interesting ... |

32 |
Advances in sensitivity encoding with arbitrary k-space trajectories
- Pruessmann, Weiger, et al.
(Show Context)
Citation Context ... to the conjugate-phase method [21]. However, for non-Cartesian k-space trajectories such as spirals, each CG-NUFFT iteration requires numerous k-space interpolations, also known as “gridding,” e.g., =-=[38]-=-. These operations are computationally expensive and poorly suited to fast hardware implementations. This paper proposes a faster iterative approach to fieldcorrected MR image reconstruction based on ... |

32 |
The gridding method for image reconstruction by Fourier transformation
- Schomberg, Timmer
(Show Context)
Citation Context ...V. SIMULATION Tl Image Display We compared four methods for field-corrected MR image reconstruction: (i) the conjugate-phase reconstruction method [5] using Voronoi-based density compensation factors =-=[52]-=- and the LS-optimal time-segmentation approximation described in §III-F, (ii) the CG-NUFFT method based on the gradient expression (12), using the time-segmentation approximation described in §III-F [... |

28 | Magnetic resonance imaging : physical principles and sequence design - Haacke - 1999 |

28 |
A homogeneity correction method for magnetic resonance imaging with time-varying gradients
- Noll, Meyer, et al.
- 1991
(Show Context)
Citation Context ...s extend readily to parallel imaging with multiple coils, e.g., [36]. The standard approach to correcting these effects is the conjugate-phase image reconstruction method and its fast variants, e.g., =-=[5]-=-, [37]. That family of methods is relatively fast since it is non-iterative, but it only partially compensates for 1soff-resonance effects. Recently, iterative methods that combine the conjugate gradi... |

27 | Unwarping of Unidirectionally Distorted EPI Images - Kybic, Thévenaz, et al. - 2000 |

24 | A technique for accurate magnetic resonance imaging in the presence of eld inhomogeneities - Chang, Fitzpatrick - 1992 |

21 |
iterative image reconstruction for MRI in the presence of field inhomogeneities
- Sutton, Noll, et al.
- 2003
(Show Context)
Citation Context ...e address the problem where the rate map z(�r) is known, i.e., where we are given relaxation maps α(�r) and field maps ω(�r) and the goal is to reconstruct the object f from the measurements y, e.g., =-=[21]-=-. For field-corrected MR reconstruction, usually one assumes that α(�r) is zero. Further applications of the general approach described here include situations where either the field map ω(�r) is unkn... |

21 |
Reconstruction of MR images from data acquired on a general non-regular grid by pseudoinverse calculation
- Walle, Barrett, et al.
- 2000
(Show Context)
Citation Context ...showing the efficiency of the proposed approach. II. REGULARIZED LS RECONSTRUCTION A. Object discretization Equation (3) is a continuous-to-discrete model that is challenging to manipulate (see [42], =-=[43]-=-). The problem is simplified by parameterizing the object f(�r) using a linear combination of N basis functions: f(�r) = N� xj p(�r − �rj) . (5) j=1 So the image reconstruction problem becomes that of... |

19 | Off-resonance correction of MR images - Schomberg - 1999 |

18 | Overview of methods for image reconstruction from projections in emission computed tomography
- Lewitt, Matej
(Show Context)
Citation Context ...sults showing the efficiency of the proposed approach. II. REGULARIZED LS RECONSTRUCTION A. Object discretization Equation (3) is a continuous-to-discrete model that is challenging to manipulate (see =-=[42]-=-, [43]). The problem is simplified by parameterizing the object f(�r) using a linear combination of N basis functions: f(�r) = N� xj p(�r − �rj) . (5) j=1 So the image reconstruction problem becomes t... |

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

15 | Compensation of susceptibility-induced BOLD sensitivity losses in echo-planar fMRI imaging - Deichmann, Josephs, et al. - 2002 |

15 |
Numerical and theoretical aspects of non-uniform sampling of bandlimited images
- Gröchenig, Strohmer
- 2001
(Show Context)
Citation Context ...ng the method more suitable for fast hardware implementations. In the absence of field inhomogeneity, this method is closely related to certain algorithms for band-limited signal interpolation, e.g., =-=[39]-=-. The Toeplitz/FFT structure has been investigated previously for MR image reconstruction in the context of sensitivity encoded imaging [40], [41]. The primary contribution here is the extension of su... |

15 |
Multifrequency interpolation for fast off-resonance correction
- Man, Pauly, et al.
- 1997
(Show Context)
Citation Context ... Instead of choosing time samples, an alternative approach is to choose a set of “frequency” samples ˇzl, for l = 1,...,L, and interpolate between these values to evaluate the exponential [37], [48], =-=[49]-=-. We can express this “frequency segmentation” approach as an approximation of the form (14) with bil � e −ˇzl(ti−¯t) , clj � cl(zj)e −zj¯t , (17) where ¯t is a nominal time reference (e.g., an echo t... |

13 |
Dynamic field map estimation using a spiral-in / spiral-out acquisition
- Sutton, Noll, et al.
- 2004
(Show Context)
Citation Context ...usually one assumes that α(�r) is zero. Further applications of the general approach described here include situations where either the field map ω(�r) is unknown and also to be estimated, e.g., [23]–=-=[25]-=-, or the relaxation map α(�r) is also to be estimated, e.g., [26], [27] or both, e.g., [28]–[35]. We focus on the case of a single receive coil, although the methods extend readily to parallel imaging... |

10 | Parsing local signal evolution directly from a single-shot MRI signal: A new approach for fMRI - Twieg - 2003 |

10 | Inhomogeneity correction using an estimated linear field map - Irarrazabal, Meyer, et al. - 1996 |

9 | Twieg, “Iterative reconstruction of singleshot spiral MRI with off-resonance - Harshbarger, B - 1999 |

9 | An inverse problem approach to the correction of distortion in EPI images - Munger, Crelier, et al. - 2000 |

9 | Conjugate phase MRI reconstruction with spatially variant sample density correction - Noll, Fessler, et al. - 2005 |

9 | Iterative image reconstruction in mri with separate magnitude and phase regularization
- Fessler, Noll
(Show Context)
Citation Context ... to the special case of quadratic regularization with an invertible Hessian, whereas the gradientbased approach that uses (12) or (13) can accommodate even non-quadratic regularization methods, e.g., =-=[54]-=-. There are several opportunities to extend this work. • When z(�r) = 0, the matrix T in (22) is Toeplitz, and good circulant preconditioners are available [47]. When z(�r) �= 1 Image and support 64 0... |

8 | Simulated phase evolution rewinding (SPHERE): A technique for reducing B inhomogeneity effects - Kadah, Hu - 1997 |

7 |
NMR imaging for magnets with large nonuniformities
- Sekihara, Matsui, et al.
- 1985
(Show Context)
Citation Context ... the BOLD effect [2], hampering the use of fMRI in brain regions near air/tissue interfaces. Numerous solutions have been proposed based both on data acquisition strategies and reconstruction methods =-=[3]-=-–[22]. In the presence of such non-Fourier effects, a more realistic model for MR measurements is the following: � E[yi] = f(�r) e − z(�r) ti e −ı2π�νi·�r d�r, (3) where ti denotes the time of the ith... |

7 | MR susceptibility misregistration correction - Sumanaweera, Glover, et al. - 1993 |

6 | Automatic field map generation and offresonance correction for projection reconstruction imaging - Nayak, Nishimura - 2000 |

6 |
Functional imaging by I0 and T ∗ 2 -parameter mapping using multiimage EPI
- Speck, Hennig
- 1998
(Show Context)
Citation Context ...bed here include situations where either the field map ω(�r) is unknown and also to be estimated, e.g., [23]–[25], or the relaxation map α(�r) is also to be estimated, e.g., [26], [27] or both, e.g., =-=[28]-=-–[35]. We focus on the case of a single receive coil, although the methods extend readily to parallel imaging with multiple coils, e.g., [36]. The standard approach to correcting these effects is the ... |

5 |
Comparison of gridding- and convolution-based iterative reconstruction algorithms for sensitivity-encoded non-Cartesian acquisition
- Eggers, Bornert, et al.
- 2002
(Show Context)
Citation Context ...gorithms for band-limited signal interpolation, e.g., [39]. The Toeplitz/FFT structure has been investigated previously for MR image reconstruction in the context of sensitivity encoded imaging [40], =-=[41]-=-. The primary contribution here is the extension of such methods to the non-Fourier model (3). Simulation results with a realistic brain field map show that the proposed CG-Toeplitz approach significa... |

4 | Iterative MR image reconstruction using sensitivity and inhomogeneity field maps
- Sutton, Fessler, et al.
- 2001
(Show Context)
Citation Context ...) is also to be estimated, e.g., [26], [27] or both, e.g., [28]–[35]. We focus on the case of a single receive coil, although the methods extend readily to parallel imaging with multiple coils, e.g., =-=[36]-=-. The standard approach to correcting these effects is the conjugate-phase image reconstruction method and its fast variants, e.g., [5], [37]. That family of methods is relatively fast since it is non... |

4 |
Reconstruction techniques for magnetic resonance imaging
- Noll
- 1991
(Show Context)
Citation Context ...end readily to parallel imaging with multiple coils, e.g., [36]. The standard approach to correcting these effects is the conjugate-phase image reconstruction method and its fast variants, e.g., [5], =-=[37]-=-. That family of methods is relatively fast since it is non-iterative, but it only partially compensates for 1soff-resonance effects. Recently, iterative methods that combine the conjugate gradient (C... |

4 |
Compensating for within-voxel susceptibility gradients
- Sutton, Noll, et al.
(Show Context)
Citation Context ...− �rj), (6) j=1 zj � α(�rj) +ıω(�rj), j = 1,...,N. (7) For cases with large within-voxel gradients of the rate map, one can use smaller voxels to reduce signal loss, albeit with increased computation =-=[44]-=- [45, p. 140]. Under these assumptions, the integral signal model (3) simplifies to the following discrete-to-discrete sum 1 : ¯yi(x) = E[yi] = Pi N� j=1 xj e −zjti e −ı2π�νi·�rj , (8) using the follo... |

3 | Efficient off-resonance correction for spiral imaging - Nayak, Tsai, et al. - 2001 |

3 | Field-corrected imaging using joint estimation of image and field map - Sutton, Fessler, et al. |

3 | EXTended Rosette ACquisition Technique (EXTRACT): a dynamic R2* mapping method using extended rosette trajectory - Lee, Noll, et al. - 2004 |

2 | Spiral sampling in magnetic resonance imaging— the effect of inhomogeneities - Yudilevich, Stark - 1987 |

2 | Simultaneous estimation of image and inhomogeneity field map
- Sutton, Noll, et al.
(Show Context)
Citation Context ...ion, usually one assumes that α(�r) is zero. Further applications of the general approach described here include situations where either the field map ω(�r) is unknown and also to be estimated, e.g., =-=[23]-=-–[25], or the relaxation map α(�r) is also to be estimated, e.g., [26], [27] or both, e.g., [28]–[35]. We focus on the case of a single receive coil, although the methods extend readily to parallel im... |

2 | A simultaneous estimation of field inhomogeneity and R2* maps using extended rosette trajectory - Lee, Fessler, et al. - 2002 |

2 | Simultaneous estimation of I0, R∗ 2 and field map using a multi-echo spiral acquisition - Sutton, Peltier, et al. |

2 |
Pruessmann, “Major speedup of reconstruction for sensitivity encoding with arbitrary trajectories
- Wajer, P
(Show Context)
Citation Context ...ain algorithms for band-limited signal interpolation, e.g., [39]. The Toeplitz/FFT structure has been investigated previously for MR image reconstruction in the context of sensitivity encoded imaging =-=[40]-=-, [41]. The primary contribution here is the extension of such methods to the non-Fourier model (3). Simulation results with a realistic brain field map show that the proposed CG-Toeplitz approach sig... |

2 | Physics-based reconstruction of magnetic resonance images - Sutton - 2003 |

2 |
Efficient iterative reconstruction for
- Barnet, Tsao, et al.
(Show Context)
Citation Context ...ality in regions with severe field inhomogeneity may enable detection of brain activation even in regions near air/tissue interfaces. An alternative CG approach has recently proposed by Barnet et al. =-=[53]-=-. That approach involves expressions of the form AA ′ , which is never Toeplitz, even when the rate map is zero, so it cannot benefit from the accelerations proposed here. Furthermore, it is limited t... |

1 | Off-resonance correction using an estimated linear time map - Akel, Rosenblitt, et al. - 2002 |

1 |
Duerk, “Block regional off-resonance correction (BRORC): A fast and effective deblurring method for spiral imaging
- Moriguchi, Dale, et al.
- 2003
(Show Context)
Citation Context ...ply ¯t = 0) and where each cl(·) denotes a frequencydomain interpolator. In the original version [37], the clj’s were chosen to be either nearest-neighbor, linear, or Hanning interpolators. (See also =-=[20]-=-.) Later, Man et al. described a least-squares approach (cf. (19) below) to choosing the interpolators cl(·) [49]. In the frequency segmentation approach, a practical issue is choosing the frequency s... |

1 |
Estimation of R∗ 2 using extended rosette acquisition
- Lee, Peltier, et al.
- 2002
(Show Context)
Citation Context ...he general approach described here include situations where either the field map ω(�r) is unknown and also to be estimated, e.g., [23]–[25], or the relaxation map α(�r) is also to be estimated, e.g., =-=[26]-=-, [27] or both, e.g., [28]–[35]. We focus on the case of a single receive coil, although the methods extend readily to parallel imaging with multiple coils, e.g., [36]. The standard approach to correc... |

1 |
New approach for estimating ∆R
- Olafsson, Noll, et al.
(Show Context)
Citation Context ...eral approach described here include situations where either the field map ω(�r) is unknown and also to be estimated, e.g., [23]–[25], or the relaxation map α(�r) is also to be estimated, e.g., [26], =-=[27]-=- or both, e.g., [28]–[35]. We focus on the case of a single receive coil, although the methods extend readily to parallel imaging with multiple coils, e.g., [36]. The standard approach to correcting t... |