Atlas-Assisted Synthetic Computed Tomography Using Deformable Image Registration

• US 20110235884A1
• Filed: 03/25/2011
• Published: 09/29/2011
• Est. Priority Date: 03/25/2010
• Status: Active Grant

• Current Assignee: Emory University
• Original Assignee: Emory University
• Inventors: Fox, Tim, Schreibmann, Eduard
• Granted Patent: US 8,774,481 B2
• Time in Patent Office: Days
• US Class Current: 382/131
• CPC Class Codes: A61B 6/037 Emission tomography
• CPC Class Codes: G06T 2207/10081 Computed x-ray tomography [CT]
• CPC Class Codes: G06T 2207/10088 Magnetic resonance imaging [MRI]
• CPC Class Codes: G06T 2207/10104 Positron emission tomography [PET]
• CPC Class Codes: G06T 2207/20016 Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
• CPC Class Codes: G06T 2207/30016 Brain
• CPC Class Codes: G06T 7/30 Determination of transform parameters for the alignment of images, i.e. image registration

Abstract

Disclosed are systems for and methods of creating a synthetic image by registering a reference or atlas image to a clinical image using both rigid and deformable image registration algorithms. In some embodiments, a synthetic computed tomography (CT) image may be created by registering an atlas CT image to a clinical image such as an MR scan. Rigid registration in some embodiments may be followed by a smoothing B-spline transform algorithm with a mutual information similarity metric and an optimizer; followed then by an image signal intensity algorithm with displacement vectors at each voxel and diffeomorphic transformations.

Claims

1. A system for creating a synthetic computed-tomography (CT) image volume, comprising:

a rigid registration algorithm configured to register a reference CT image volume to a first clinical image volume and thereby produce a first registration;

a smoothing function configured to further refine said first registration and thereby produce a smoothed registration; and

a voxelwise intensity algorithm configured to further refine said smoothed registration and thereby produce a synthetic CT image volume having characteristics correlated to those of said first clinical image volume.

2. The system of claim 1, further comprising:

a map of attenuation coefficients based on said synthetic CT image volume; and

an attenuation correction engine configured to correct the attenuation in a second clinical image volume based on said map.

3. The system of claim 1, wherein said rigid registration algorithm is further configured to register said first clinical image volume to each of a plurality of candidate CT image volumes and thereby obtain a plurality of corresponding candidate similarity metrics;

and wherein said system further comprises;

a selection tool configured to select a reference CT image volume from among said plurality of candidate CT image volumes based on a comparison of each of said plurality of corresponding candidate similarity metrics to a desired similarity metric.

4. The system of claim 1, wherein said rigid registration algorithm comprises:

an affine transform algorithm having a plurality of variables including translation, rotation, scale, and shear;

a similarity metric representing the alignment of said reference CT image volume to said first clinical image volume;

an optimizer configured to iteratively adjust said variables until said similarity metric achieves a minimum limit.

5. The system of claim 4, wherein said optimizer further comprises:

an iteration maximum;

a similarity metric minimum; and

a module configured to halt said optimizing after either said iteration maximum equals about thirty or said similarity metric minimum falls below about 10<sup>−

5</sup>.

6. The system of claim 1, wherein said smoothing function comprises:

a B-spline transform wherein a deformation is defined by a grid of nodes superimposed on said first registration, said transform having a plurality of variables including node deformation;

a similarity metric representing the alignment of said first registration to said first clinical image volume; and

an optimizer configured to iteratively adjust said variables until said similarity metric achieves a minimum limit.

7. The system of claim 6, wherein said B-spline transform comprises:

a first stage B-spline transform wherein said grid of nodes comprises a first grid of nodes arrayed at a first spacing distance superimposed on said first registration, said first stage producing an intermediate registration; and

a second stage B-spline transform wherein said grid of nodes further comprises a second grid of nodes arrayed at a second spacing distance superimposed on said intermediate registration, wherein said second spacing distance is shorter than said first spacing distance.

8. The system of claim 1, wherein said voxelwise algorithm comprises:

a deformable image registration transform having a plurality of nonlinear variables;

a displacement vector defined at least at each voxel;

a similarity metric representing the alignment of said smoother registration to said first clinical image volume; and

an optimizer configured to iteratively adjust said nonlinear variables according to said displacement vector until said similarity metric achieves a minimum limit.

9. The system of claim 8, wherein said optimizer is further configured to iteratively adjust said nonlinear variables according to a partial differential equation having a diffeomorphic update step.

10. The system of claim 8, wherein said similarity metric comprises a variational expression of a mutual information component that is varied according to a local value of an image intensity gradient based at least partially on said displacement vector.

11. A method of creating a synthetic computed-tomography (CT) image volume, the method comprising:

obtaining a first clinical image volume;

retrieving a reference CT image volume;

applying a rigid registration algorithm to register said reference CT image volume to said first clinical image volume and thereby produce a first registration;

applying a smoothing function to said first registration and thereby produce a smoothed registration; and

applying a voxelwise intensity algorithm to said smoothed registration and thereby produce a synthetic CT image volume having characteristics correlated to those of said first clinical image volume.

12. The method of claim 11, further comprising:

obtaining a second clinical image volume;

deriving a map of attenuation coefficients based on said synthetic CT image volume; and

correcting said second clinical image volume for attenuation based on said map.

13. The method of claim 11, wherein said step of retrieving a reference CT image volume comprises:

establishing a desired similarity metric;

selecting a plurality of candidate CT image volumes;

executing said rigid registration algorithm to register said first clinical image volume to each of said plurality of candidate CT image volumes and thereby obtain a plurality of corresponding candidate similarity metrics; and

selecting said a reference CT image volume from among said plurality of candidate CT image volumes based on a comparison of each of said plurality of corresponding candidate similarity metrics to a desired similarity metric.

14. The method of claim 11, wherein said step of applying a rigid registration algorithm comprises:

executing an affine transform algorithm having a plurality of variables including translation, rotation, scale, and shear;

selecting a similarity metric representing the alignment of said reference CT image volume to said first clinical image volume; and

optimizing said variables iteratively until said similarity metric achieves a minimum limit.

15. The method of claim 14, wherein said step of optimizing further comprises:

setting an iteration maximum;

setting a similarity metric minimum; and

halting said optimizing after either said iteration maximum equals about thirty or said similarity metric minimum falls below about 10<sup>−

5</sup>.

16. The method of claim 11, wherein said step of applying a smoothing function comprises:

executing a B-spline transform wherein a deformation is defined by a grid of nodes superimposed on said first registration, said transform having a plurality of variables including node deformation;

selecting a similarity metric representing the alignment of said first registration to said first clinical image volume; and

optimizing said variables iteratively until said similarity metric achieves a minimum limit.

17. The method of claim 16, wherein said step of executing a B-spline transform further comprises:

executing a first stage B-spline transform wherein said grid of nodes comprises a first grid of nodes arrayed at a first spacing distance superimposed on said first registration, said first stage producing an intermediate registration; and

executing a second stage B-spline transform wherein said grid of nodes further comprises a second grid of nodes arrayed at a second spacing distance superimposed on said intermediate registration, wherein said second spacing distance is shorter than said first spacing distance.

18. The method of claim 11, wherein said step of executing a voxelwise algorithm comprises:

executing a deformable image registration transform having a plurality of nonlinear variables;

defining at least a displacement vector at each voxel;

selecting a similarity metric representing the alignment of said smoother registration to said first clinical image volume; and

optimizing said nonlinear variables iteratively, according to said displacement vector, until said similarity metric achieves a minimum limit.

19. The method of claim 18, wherein said step of optimizing further comprises:

optimizing said nonlinear variables according to a partial differential equation having a diffeomorphic update step.

20. The method of claim 18, wherein said step of selecting further comprises:

selecting a similarity metric having a variational expression of a mutual information component that is varied according to a local value of an image intensity gradient based at least partially on said displacement vector.

21. A non-transitory computer-readable medium, the contents of which cause a computing system to perform a method of creating a synthetic computed-tomography (CT) image volume, the method comprising:

obtaining a first clinical image volume;

retrieving a reference CT image volume;

applying a rigid registration algorithm to register said reference CT image volume to said first clinical image volume and thereby produce a first registration;

applying a smoothing function to said first registration and thereby produce a smoothed registration; and

applying a voxelwise intensity algorithm to said smoothed registration and thereby produce a synthetic CT image volume having characteristics correlated to those of said first clinical image volume.

22. The computer-readable medium of claim 21, wherein said method further comprises:

obtaining a second clinical image volume;

deriving a map of attenuation coefficients based on said synthetic CT image volume; and

correcting said second clinical image volume for attenuation based on said map.

23. The computer-readable medium of claim 21, wherein said step of applying a rigid registration algorithm comprises:

executing an affine transform algorithm having a plurality of variables including translation, rotation, scale, and shear;

selecting a similarity metric representing the alignment of said reference CT image volume to said first clinical image volume; and

optimizing said variables iteratively until said similarity metric achieves a minimum limit.

24. The computer-readable medium of claim 21, wherein said step of applying a smoothing function comprises:

executing a B-spline transform wherein a deformation is defined by a grid of nodes superimposed on said first registration, said transform having a plurality of variables including node deformation;

selecting a similarity metric representing the alignment of said first registration to said first clinical image volume; and

optimizing said variables iteratively until said similarity metric achieves a minimum limit.

25. The computer-readable medium of claim 21, wherein said step of executing a voxelwise algorithm comprises:

executing a deformable image registration transform having a plurality of nonlinear variables;

defining at least a displacement vector at each voxel;

selecting a similarity metric representing the alignment of said smoother registration to said first clinical image volume; and

optimizing said nonlinear variables iteratively, according to said displacement vector, until said similarity metric achieves a minimum limit.