Method and system for registration of contrast-enhanced images with volume-preserving constraint
View Patent ↗A method and system for registering a first image of, for example, a liver and a second image of the liver being contrast-enhanced comprises: deriving a statistical similarity measure between images; deriving a smooth divergence-free vector field derived from a gradient of the statistical similarity measure; and integrating the vector field for providing a fluid-based algorithm including a volume-preserving constraint for a transformation for registering the images.
1. A method for registering images, comprising:
(a) acquiring a reference image and a study image;
(b) setting a deformation field to an initial deformation field;
(c) deriving a similarity measure from said reference and study images and said initial deformation field;
(d) computing a displacement field from said similarity measure;
(e) smoothing said displacement field to derive a regularized displacement field;
(f) applying a volume constraint to said regularized displacement field to derive a constrained displacement field;
(g) utilizing said constrained displacement field and said initial deformation field for deriving an updated deformation field;
(h) deriving an updated similarity measure value from said reference and study images and said updated deformation field;
(i) computing an updated displacement field from said updated similarity measure;
(j) smoothing said updated displacement field to derive an updated regularized displacement field;
(k) applying a volume constraint to said updated regularized displacement field to derive an updated constrained displacement field;
(l) utilizing said updated constrained displacement field and said updated deformation field for deriving a further updated deformation field; and
(m) iteratively performing steps (h) through (m) until an end condition is met, whereupon ending iteration and outputting a furthest updated deformation field.
2. A method for registering images as recited in claim 1 , wherein:
said step (b) for setting a deformation field to an initial deformation field comprises setting said deformation field to zero.
3. A method for registering images as recited in claim 1 , and for performing steps of said method in a coarse-to-fine framework:
in a first performance in a relatively coarse resolution;
in subsequent performances in respective progressively finer resolutions;
wherein, for said first performance, said step (b) comprises setting said deformation field to zero; and
wherein, for each of said subsequent performances, said step (b) comprises setting said deformation field to a respective updated deformation field from the immediately preceding performance.
4. A method for registering images as recited in claim 1 , wherein:
said step (c) for deriving a similarity measure value comprises deriving the sum of squared differences, local cross-correlation, and mutual information for said images.
5. A method for registering images as recited in claim 1 , wherein:
said step (d) for computing a displacement field comprises:
deriving the gradient of said similarity measure, and
determining said displacement field as following said gradient.
6. A method for registering images as recited in claim 1 , wherein:
said step (e) for smoothing said displacement field comprises utilizing a fast recursive filtering technique approximating a Gaussian smoothing.
7. A method for registering images as recited in claim 1 , wherein:
said step (f) for applying a volume constraint comprises applying a substantially divergence-free constraint upon said regularized displacement field comprising solving a Poisson equation using a multi-grid method.
8. A method for registering images as recited in claim 1 , wherein:
said step (g) for utilizing said constrained displacement field and said initial deformation field for deriving an updated deformation field comprises using a control parameter in an updating step to ensure that the extent of updating does not exceed one voxel.
9. A method for registering images as recited in claim 1 , wherein said end condition comprises ending when a given maximum number of iterations has been performed.
10. A method for registering images as recited in claim 9 , wherein said end condition comprises ending when a similarity cost stops decreasing.
11. A computerized system for registering a first digital image of an organ and a second digital image of said organ being contrast-enhanced comprising:
a memory device for storing a program and other data; and
a processor in communication with said memory device, said processor being operative with said program to perform:
(a) acquiring a reference image and a study image;
(b) setting a deformation field to an initial deformation field;
(c) deriving a similarity measure from said reference and study images and said initial deformation field;
(d) computing a displacement field from said similarity measure;
(e) smoothing said displacement field to derive a regularized displacement field;
(f) applying a volume constraint to said regularized displacement field to derive a constrained displacement field;
(g) utilizing said constrained displacement field and said initial deformation field for deriving an updated deformation field;
(h) deriving an updated similarity measure value from said reference and study images and said updated deformation field;
(i) computing an updated displacement field from said updated similarity measure;
(j) smoothing said updated displacement field to derive an updated regularized displacement field;
(k) applying a volume constraint to said updated regularized displacement field to derive an updated constrained displacement field;
(l) utilizing said updated constrained displacement field and said updated deformation field for deriving a further updated deformation field; and
(m) iteratively performing steps (h) through (m) until an end condition is met, whereupon ending iteration and outputting a furthest updated deformation field.
12. A system for registering images as recited in claim 11 , and for operating said system in a coarse-to-fine framework:
in a first operation in a relatively coarse resolution;
in subsequent operations in respective progressively finer resolutions;
wherein, for said first operation, said setting a deformation field to an initial deformation field comprises setting said deformation field to zero; and
wherein, for each of said subsequent operations, said setting a deformation field to an initial deformation field comprises setting said deformation field to a respective updated deformation field from the immediately preceding operation.
13. A system for registering images as recited in claim 11 , wherein:
said deriving a similarity measure value comprises deriving the sum of squared differences, local cross-correlation, and mutual information for said images.
14. A system for registering images as recited in claim 11 , wherein:
said computing a displacement field comprises:
deriving the gradient of said similarity measure, and
determining said displacement field as following said gradient.
15. A system for registering images as recited in claim 11 , wherein:
said smoothing said displacement field comprises utilizing fast recursive Gaussian filtering.
16. A system for registering images as recited in claim 11 , wherein:
said applying a volume constraint comprises applying a substantially divergence-free constraint upon said regularized displacement field, comprising solving a Poisson equation with a multi-grid method.
17. A system for registering images as recited in claim 11 , wherein:
said utilizing said constrained displacement field and said initial deformation field for deriving an updated deformation field comprises using a control parameter in an updating operation to ensure that the extent of updating does not exceed one voxel.