Apparatus and method for correcting susceptibility artefacts in a magnetic resonance image

Patent metadata

Field	Value
Publication number	US-9702955-B2
Application number	US-201414764094-A
Country	US
Kind code	B2
Filing date	Jan 31, 2014
Priority date	Feb 1, 2013
Publication date	Jul 11, 2017
Grant date	Jul 11, 2017

Abstract

An apparatus and method are provided for performing phase unwrapping for an acquired magnetic resonance (MR) image. The method includes modelling the MR phase in the MR image using a Markov random field (MRF) in which the true phase φ(t) and the wrapped phase φ(w) are modelled as random variables such that at voxel i of said MR image φ(t)(i)=φ(w)(i)+2πn(i), where n(i) is an unknown integer that needs to be estimated for each voxel i. The method further includes constructing a graph consisting of a set of vertices V and edges E and two special terminal vertices representing a source s and sink t, where there is a one-to-one correspondence between cuts on the graph and configurations of the MRF, a cut representing a partition of the vertices V into disjoint sets S and T such that sεS and tεT. The method further includes finding the minimum energy configuration, E(n(i)|φ(w)) of the MRF on the basis that the total cost of a given cut represents the energy of the corresponding MRF configuration, where the cost of a cut is the sum of all edges going from S to T across the cut boundary. The method further includes using the values of n(i) in the minimum energy configuration to perform the phase unwrapping from φ(w) to φ(t) for the MR image. A confidence may be computed for each voxel using dynamic graph cuts. The unwrapped phase from two MR images acquired at different times may be used to estimate a field map from the phase difference between the two MR images. The field map may be converted into a deformation field which is then used to initialize a non-rigid image registration of the acquired MR image against a reference image. The deformation field of the non-rigid registration is controlled to be smoother where the confidence is high.

First claim

What is claimed is:

1. A method of correcting susceptibility artefacts in an acquired magnetic resonance image comprising the steps of:
performing phase unwrapping for an acquired magnetic resonance (MR) image, including computing a confidence for the phase unwrapping for each voxel of the acquired MR image;
generating a field map for the acquired magnetic resonance image from the unwrapped phase;
converting the field map to a deformation field; and
using the deformation field to initialise a non-rigid image registration of the acquired MR image against a reference image, wherein the deformation field of the non-rigid registration is controlled to be smoother where said confidence is high.

2. The method of claim 1 , wherein performing the phase unwrapping for an acquired magnetic resonance (MR) image comprises:
modelling the MR phase in the MR image using a Markov random field (MRF) in which the true phase φ(t) and the wrapped phase φ(w) are modelled as random variables such that at voxel i of said MR image φ(t)(i)=φ(w)(i)+2πn(i), where n(i) is an unknown integer that needs to be estimated for each voxel i;
constructing a graph consisting of a set of vertices V and edges E and two special terminal vertices representing a source s and sink t, where there is a one-to-one correspondence between cuts on the graph and configurations of the MRF, a cut representing a partition of the vertices V into disjoint sets S and T such that sεS and tεT;
finding the minimum energy configuration, E(n(i)|φ(w)) of the MRF on the basis that the total cost of a given cut represents the energy of the corresponding MRF configuration, where the cost of a cut is the sum of all edges going from S to T across the cut boundary; and
using the values of n(i) in the minimum energy configuration to perform the phase unwrapping from φ(w) to φ(t) for the MR image.

3. The method of claim 2 , wherein the MR phase is modelled as a piecewise smooth function where the smooth component is due to the inhomogeneities in the static MR field and the non-smooth component arises due to changes in magnetic susceptibility from one tissue type to another, and wherein the spatial smoothness is enforced by modelling the true phase as the MRF and a smoothness model is incorporated by using a potential function.

4. The method of claim 3 , wherein the true phase is modelled as a six-neighbourhood pairwise MRF where a convex pairwise potential function is defined between immediately adjacent neighbouring voxels.

5. The method of claim 4 , wherein the convex pairwise potential function is sum of squared differences.

6. The method of claim 2 , further comprising weighting voxel contributions on the basis that phase measurements in low signal areas of the MR image tend to be less reliable, such that phase measurements in high signal areas of the MR image are assigned a higher weight that phase measurements in low signal areas of the MR image.

7. The method of claim 6 , wherein the weight assigned to a given voxel corresponds to the probability that the contribution of noise to the phase of a given voxel is less than ε radians, where ε is set to be a small value greater than zero.

8. The method of claim 7 , further comprising selecting a region in the MR image containing only air to determine the noise variance of the MR image, wherein said noise variance is then used for determining the weight to be assigned to a given voxel.

9. The method of claim 2 , wherein computing a confidence comprises performing a confidence estimation at each voxel for the phase unwrapping.

10. The method of claim 9 , wherein the confidence associated with a given label for a random variable is estimated as the ratio of the min-marginal energy associated with that labelling to the sum of the min-marginal energies for that random variable for all possible labels, wherein the min-marginal is the minimum energy obtained when a random variable is constrained to take a certain label and the minimisation is performed over all remaining variables.

11. The method of claim 10 , wherein the confidence is computed for each voxel using dynamic graph cuts.

12. The method of claim 11 , wherein using dynamic graph cuts comprises constraining a given MRF node to belong to the source or sink by adding an infinite capacity edge between the given MRF node and the respective terminal node, and computing the min-marginal energies by optimizing the resulting MRF.

13. The method of claim 1 , wherein the confidence is computed for each voxel using one of the following techniques: Expectation Propagation, Variational Bayes, Markov Chain Monte Carlo (MCMC) and Hamiltonian MCMC simulations.

14. The method of claim 1 , wherein generating the field map comprises using the unwrapped phase from two MR images or the unwrapped phase difference between the two MR images to estimate a field map.

15. The method of claim 14 , further comprising converting the field map into a deformation field.

16. The method of claim 15 , wherein the deformation field is used to initialise a non-rigid image registration of the acquired MR image against a reference image.

17. The method of claim 16 , wherein the deformation field for performing the non-rigid image registration is parameterised using cubic B-splines, where the B-splines are constrained to move only in the phase encode direction.

18. The method of claim 17 , wherein a similarity measure used for the non-rigid image registration is based on local information which encodes spatial context by varying the contribution of voxels according to their spatial coordinates.

19. The method of claim 18 , wherein the non-rigid image registration is represented as a discrete set of displacements in the phase encode direction, and the non-rigid image registration is modelled as a first-order Markov random field.

20. The method of claim 19 , wherein the first derivative of the deformation field for the non-rigid image registration is penalised to ensure a smooth transformation.

21. The method of claim 20 , wherein a penalty term for penalising the first derivative of the deformation field is modulated by the confidence computed for each voxel, such that the penalty weight is higher in regions of high confidence than in regions of low confidence.

22. The method of claim 1 , wherein a confidence for each voxel of the field map is determined from the computed confidence for the phase unwrapping and an estimate of uncertainty in the MR image, and wherein the confidence of the field map is used to control the smoothness of the deformation field.

23. A method of performing phase unwrapping for an acquired magnetic resonance (MR) image comprising:
modelling the MR phase in the MR image using a Markov random field (MRF) in which the true phase φ(t) and the wrapped phase φ(w) are modelled as random variables such that at voxel i of said MR image φ(t)(i)=φ(w)(i)+2πn(i), where n(i) is an unknown integer that needs to be estimated for each voxel i;
constructing a graph consisting of a set of vertices V and edges E and two special terminal vertices representing a source s and sink t, where there is a one-to-one correspondence between cuts on the graph and configurations of the MRF, a cut representing a partition of the vertices V into disjoint sets S and T such that sεS and tεT;
finding the minimum energy configuration, E(n(i)|φ(w)) of the MRF on the basis that the total cost of a given cut represents the energy of the corresponding MRF configuration, where the

Assignees

• Ucl Business Plc

Inventors

• Ourselin Sebastien
• Daga Pankaj

Classifications

• G01R33/56545

  caused by finite or discrete sampling, e.g. Gibbs ringing, truncation artefacts, phase aliasing artefacts · CPC title

• G01R33/565 Primary

  Correction of image distortions, e.g. due to magnetic field inhomogeneities · CPC title

• G01R33/24

  for measuring direction or magnitude of magnetic fields or magnetic flux · CPC title

• G01R33/5608

  Data processing and visualization specially adapted for MR, e.g. for feature analysis and pattern recognition on the basis of measured MR data, segmentation of measured MR data, edge contour detection on the basis of measured MR data, for enhancing measured MR data in terms of signal-to-noise ratio by means of noise filtering or apodization, for enhancing measured MR data in terms of resolution by means for deblurring, windowing, zero filling, or generation of gray-scaled images, colour-coded images or images displaying vectors instead of pixels (image data processing or generation, in general G06T) · CPC title

• G01R33/56536 Primary

  due to magnetic susceptibility variations · CPC title