Three-dimensional image registration method for spiral CT angiography

• US 5,839,440 A
• Filed: 06/17/1994
• Issued: 11/24/1998
• Est. Priority Date: 06/17/1994
• Status: Expired due to Term

• Current Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Original Assignee: Siemens Corporate Research Incorporated (Siemens AG)
• Inventors: Yeo, Boon-Lock, Yeung, Minerva M., Liou, Shih-Ping
• Primary Examiner(s): Smith, Ruth S.
• Application Number:
  US08/261,643
• Time in Patent Office: 1,621 Days
• Field of Search: 128/653.1, 128/653.2, 128/654, 128/653.4, 364/413.13, 364/413.14, 364/413.23, 364/98.12, 378/4, 378/901, 395/119, 395/924
• US Class Current: 600/431
• CPC Class Codes: A61B 6/5264 due to motion
• CPC Class Codes: G06T 2207/30004 Biomedical image processing
• CPC Class Codes: G06T 7/33 using feature-based methods
• CPC Class Codes: G06T 7/38 Registration of image sequences
• CPC Class Codes: Y10S 378/901 Computer tomography program or processor

Abstract

A three-dimensional image registration method also uses an algorithm and applies it to the solution of such problems in 3D CT DSA. The method can deal with incomplete or partial volumetric data in registration and correct patient global motion prior to subtraction even when it is coupled with local unconscious/nonrigid movements. Experimental results demonstrate the effectiveness of this algorithm on several clinical spiral CT data.

Claims

1. A computer-implemented method for three-dimensional image registration in an imaging technique utilizing mask images, and respective data thereto, obtained before opacification and contrast images acquired after the injection of a contrast media bolus, said method comprising the steps of:

resampling serial axial CT mask and contrast images into respective isotropic 3D volumes;

selecting 3D feature points in said mask volume;

establishing correspondences in said contrast volume;

processing resulting sparse 3D image flow vectors by an iterative random algorithm and computing motion parameters, translation and rotation, in a least squares optimized sense that are agreed upon by at least a preset percentage of pairs whereby patient motion is found;

after patient motion is found, transforming said mask volume accordingly and subtracting it from said contrast volume; and

rendering and displaying a resulting volume.

2. A computer-implemented method in accordance with claim 1 wherein said step of resampling serial axial CT mask and contract images comprises:

performing trilinear interpolation for inter-slice voxels; and

performing subsampling for voxels in orthogonal directions.

3. A computer-implemented method in accordance with claim 1 wherein said step of resampling serial axial CT mask and contrast images reduces image data such that an image registration algorithm can achieve a level of computational efficiency desired in a computed tomography angiography CTA application.

4. A computer-implemented method in accordance with claim 1 wherein said image registration algorithm uses only subsampled data associated with points of interest subsampled in said step of performing subsampling to locate 3D feature points of interest.

5. A computer-implemented method in accordance with claim 4 wherein after said points of interest have been located, said algorithm discards said subsampled data.

6. A computer-implemented method in accordance with claim 5 wherein after said subsampled data has been discarded, said algorithm refers back to said first mentioned, original image in a subsequent registration process.

7. A computer-implemented method for three-dimensional image registration in an imaging technique utilizing mask images, and respective data related thereto, obtained before opacification and contrast images, and respective data related thereto, acquired after the injection of a contrast media bolus, said method comprising the steps of:

resampling serial axial CT mask and contrast images into respective isotropic 3D volumes;

selecting 3D feature points in said mask volume;

establishing correspondences in said contrast volume;

processing resulting sparse 3D image flow vectors by an iterative random algorithm as follows;

(a) start with a desired maximum individual residual error ε and

desired size s;

(b)select randomly a small number of points, N_i, from a set of points in said mask volume, V to form S and the corresponding points from a set of points in contrast volume, W to form S'"'"';

(c) compute a rotation matrix R and a translation vector t for a set S and a set S'"'"' using unit quaternions and if the maximum individual residual error is greater than ε

, repeat this step until such is no longer the case;

(d) randomly pick a fixed number of new points from the remaining ones in V and W, and compute new transform parameters and, if the error constraint is again satisfied, append these points to S and S'"'"' and if not, repeat this step; and

(e) repeat step (d) until the size of S and S'"'"' is greater than or equal to s, at which point terminate with V'"'"'←

S and W'"'"'←

S'"'"', or restart with step (b) if, after a predetermined number of times, T1, of repeating step (d), the size of S does not reach s, or restart with step (b) with a new ε

if for the given ε

, the desired size s is not obtained after a predetermined number of restarts, T2.

8. A computer-implemented method in accordance with claim 7 wherein after said rotation matrix and a translation vector are computed, a geometric transformation is carried out.

9. A computer-implemented method in accordance with claim 8 wherein said mask volume is transformed according to new transform parameters and resampled to an isotropic volume for display purposes.

10. A computer-implemented method in accordance with claim 9 wherein said contrast volume is also resampled to an isotropic volume of the same dimension.

11. A computer-implemented method in accordance with claim 10 wherein sub-voxel intensity values are computed by trilinear interpolation.

12. A computer-implemented method in accordance with claim 11 including a rendering step wherein subtraction of resampled contrast volume by the geometrically transformed and resampled volume is performed to yield a subtracted DSA volume.

13. A computer-implemented method in accordance with claim 12 wherein said rendering step is performed using a known type of renderer.

14. A computer-implemented method in accordance with claim 13 wherein said rendering step is performed so as to result in a variable number of images each of which corresponds to a projection at a different viewing angle.

15. A computer-implemented method in accordance with claim 14 wherein rendered results are reviewed using an animator program for providing a fast display of successive frames.

24. A computer-implemented method for three-dimensional image registration in accordance with claim 7, wherein step (c) of computing motion parameters R and t is performed in accordance with the following steps:

compute R and t subject to the minimization of the sums of squares of errors ##EQU13## utilizing a quaternion based algorithm, wherein a unit quaternion is a four vector q_R = q₀ q₁ q₂ q₃ !^t where q₀ ≧

0, and q₀² +q₁² +q₂² +q₃² =1 and is way to represent a unit rotation, R being computed as;

##EQU14## compute centroids μ

_v V of the set V and μ

_w of the set W in accordance with ##EQU15## and ##EQU16## compute a 3×

3 cross-covariance matrix M, ##EQU17## whereof the elements are the sums of products of coordinates in V and W, with M being capable of being written in the form ##EQU18## where ##EQU19## and so on, and where v_i = X_v,i Y_v,i Z_v,i !^t and w_i = X_w,i Y_w,i Z_w,i !^t ; and

compute an optimal translation vector

space="preserve" listing-type="equation">t=μ

.sub.w -Rμ

.sub.v.

where the unit quaternion of interest is the unit eigenvector corresponding to the most positive eigenvalue of the 4×

4 symmetric matrix N, where ##EQU20##

16. A computer-implemented method for three-dimensional image registration in an imaging technique utilizing mask images, and respective data related thereto, obtained before opacification and contrast images, and respective data related thereto, acquired after the injection of a contrast media bolus, said method comprising the steps of:

resampling serial axial CT mask and contrast images into respective isotropic 3D volumes;

selecting a set of 3D feature points in said mask volume;

establishing correspondence with a set of points in said contrast volume;

processing resulting sparse 3D image flow vectors by an iterative random algorithm as follows;

(a) start with a desired maximum individual residual error ε and

desired size s;

(b)select randomly a small number of points, N_i, from a set of points in said mask volume, V to form S and the corresponding points from a set of points in said contrast volume, W to form S'"'"';

(c) compute a rotation matrix R and a translation vector t for a set S and a set S'"'"' using unit quaternions, wherein a unit quaternion expresses a unit rotation, and if the maximum individual residual error is greater than ε

, repeat this step until such is no longer the case;

(d) randomly pick a fixed number of new points from the remaining ones in V and W, and compute new transform parameters and, if the error constraint is again satisfied, append these points to S and S'"'"' and if not, repeat this step; and

(e) repeat step (d) until the size of S and S'"'"' is greater than or equal to s, at which point terminate with V'"'"'←

S and W'"'"'←

S'"'"', or restart with step (b) if, after a predetermined number of times, T1, of repeating step (d), the size of S does not reach s, or restart with step (b) with a new ε

if for the given ε

, the desired size s is not obtained after a predetermined number of restarts, T2.

17. A computer-implemented method in accordance with claim 16 wherein in step (c) a rotation matrix and a translation vector are obtained for transforming said mask volume to said contrast volume.

18. A computer-implemented method in accordance with claim 17 wherein said rotation matrix and said translation vector are computed subject to a minimization of sums of squares of errors.

19. A computer-implemented method in accordance with claim 18 wherein respective centroids of said rotation matrix and said translation vector are obtained.

20. A computer-implemented method in accordance with claim 19 wherein a cross-covariance matrix is computed, said cross-covariance matrix having elements which are respective sums of products of coordinates in said rotation matrix and said translation vector.

21. A computer-implemented method in accordance with claim 20 wherein said unit quaternions are obtained as the unit eigenvector corresponding to a most positive eigenvalue of a symmetric matrix.

22. A computer-implemented method in accordance with claim 21 wherein said symmetric matrix is a 4×

4 matrix whereof elements are said respective sums of products of coordinates in said rotation matrix and said translation vector.

23. A computer-implemented method in accordance with claim 22 wherein an optimal translation vector is then derived as the difference between (A) said centroid of said set of points in said contrast volume, and (B) the product of said centroid of said set of 3D feature points in said mask volume and said rotation matrix.