Image similarity-based finite element model registration

Patent metadata

Field	Value
Publication number	US-9262583-B2
Application number	US-201414225850-A
Country	US
Kind code	B2
Filing date	Mar 26, 2014
Priority date	Mar 29, 2013
Publication date	Feb 16, 2016
Grant date	Feb 16, 2016

Abstract

Apparatus, methods, and other embodiments associated with evaluating global deformations and local deformations in a prostate are described. One example apparatus includes logics that evaluate global and local deformations in a prostate and register a pre-External Beam Radiation Treatment (EBRT) three dimensional (3D) magnetic resonance (MR) image with a post-EBRT 3D MR image. An image acquisition logic acquires a pre-EBRT image and a post EBRT image of an organ, item, or volume. An image texture information logic extracts image texture information from the pre-EBRT and post-EBRT images. A finite element model (FEM) transformation logic constructs a FEM of the volume imaged in the pre-EBRT image, deforms the FEM, deforms the pre-EBRT image as a function of the deformed FEM, and maximizes the image texture similarity between the deformed pre-EBRT image and the post-EBRT image. A registration logic registers the pre-EBRT image with the post-EBRT image based on the transformation.

First claim

What is claimed is:

1. A non-transitory computer-readable storage medium storing computer executable instructions that when executed by a computer cause the computer to perform a method associated with evaluating global deformations and local deformations in a prostate due to effects of External Beam Radiation Treatment (EBRT), the method comprising:
accessing a three dimensional (3D) pre-EBRT magnetic resonance (MR) image of a prostate, where the 3D pre-EBRT image comprises a first set of voxels and first image texture information about the first set of voxels;
accessing a 3D post-EBRT MR image of the prostate, where the 3D post-EBRT image comprises a second set of voxels and second image texture information about the second set of voxels;
extracting the first image texture information from the 3D pre-EBRT image and extracting the second image texture information from the 3D post-EBRT image;
constructing a finite element model (FEM) of the 3D pre-EBRT image, where the FEM discretizes structures of the prostate into elements of the FEM, where the elements of the FEM represent a geometric shape, where the elements of the FEM contain physical properties including elasticity or compressibility, and where the elements of the FEM are connected at nodes of the FEM;
producing a deformation of the FEM by applying a series of non-zero external forces to the surface of the prostate represented by the FEM;
calculating a transformation of the FEM, based at least in part, on the deformation, where the transformation is parameterized based on the series of non-zero external forces applied to the surface of the prostate;
producing a deformed first image texture by deforming, as a function of the deformation of the FEM, the first image texture;
determining, as a function of the transformation, a non-zero external force that maximizes an image texture similarity between the deformed first image texture and the second image texture; and
upon determining the non-zero external force that maximizes the image texture similarity between the deformed first image texture and the second image texture, registering the 3D pre-EBRT image to the 3D post-EBRT image based, at least in part, on the transformation, where registering the 3D pre-EBRT image to the 3D post-EBRT image comprises spatially aligning the 3D pre-EBRT image with the 3D post-EBRT image.

2. The non-transitory computer-readable storage medium of claim 1 , where the FEM is a 3D FEM, where the 3D FEM comprises a plurality of nodes N, where elements of the 3D FEM represent one of a tetrahedron or a hexahedron, where the tetrahedron is defined by four corner nodes, and where the hexahedron is defined by eight corner nodes.

3. The non-transitory computer-readable storage medium of claim 2 , where interaction of nodes in three dimensions is defined by a 3N×3N matrix K, where K is a sparse, symmetrical matrix, and where a 3N×1 vector V represents co-ordinates of nodes in three dimensions, N being an integer.

4. The non-transitory computer-readable storage medium of claim 3 , where the series of non-zero external forces is denoted by M, where a 3N×1 vector F represents the series of non-zero external forces M applied to a node in three dimensions, where M is less than or equal to the number of nodes N.

5. The non-transitory computer-readable storage medium of claim 4 , where a 3N×1 vector U represents displacement of nodes in three dimensions by the series of non-zero external forces.

6. The non-transitory computer-readable storage medium of claim 5 , where displacement of nodes in the FEM by the series of non-zero external forces is computed according to U=K −1 ·F, where U is estimated by minimizing ∥F−K·U∥ 2 .

7. The non-transitory computer-readable storage medium of claim 6 , where the transformation is defined from the displacement of nodes, where nodes S c surrounding a voxel c in the first set of voxels are defined by corners of an element of the FEM, where the element contains the voxel c.

8. The non-transitory computer-readable storage medium of claim 7 , where a transformation of the voxel c has 3M degrees of freedom, where the transformation of the voxel c is defined as a linear interpolation of the displacement of the nodes S c , where the transformation of the voxel c is expressed as

T
⁡

(
c
)

=

c
+

∑

n
∈

S
c

⁢



c
-

v
n



2

·

u
n

∑

n
∈

S
c

⁢



c
-

v
n



2

,

where u n denotes displacement of a node n, and where v n denotes coordinates of a node n.

9. The non-transitory computer-readable storage medium of claim 8 , where registering the 3D pre-EBRT image to the 3D post-EBRT image comprises calculating T=arg max E(C Post ,T(C pre )), where T(C Pre )={C Pre ,f Pre (T(c))}, where C pre is the 3D pre-EBRT image, C Post is the 3D post-EBRT image, f Pre is image texture information for a voxel, and where E denotes an image energy, where the image energy is defined as mutual information between a texture transformation of the 3D pre-EBRT image and the 3D post-EBRT image.

10. The non-transitory computer-readable storage medium of claim 9 , the method comprising using a particle swarm optimizer to determine a series of non-zero external forces that maximize E.

11. The non-transitory computer-readable storage medium of claim 1 , where producing a deformation of the FEM comprises applying a series of non-zero internal forces to internal structures of the prostate represented by the FEM.

12. A non-transitory computer-readable storage medium storing com

Assignees

• Univ Case Western Reserve

Inventors

• Madabhushi Anant
• Toth Robert

Classifications

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06T7/0014 Primary

  using an image reference approach · CPC title

• G06T7/0032

  Physics · mapped topic

• G06F19/321 Primary

  Physics · mapped topic

• G06K9/4671

  Physics · mapped topic