What is claimed is:

1. A method for registering a first volumetric image (110) and a second volumetric image (112) , each image comprising a three-dimensional array of gray scale voxel values, the method comprising:

(a) defining mutation probabilities (316, 340) for a plurality of aligned pairs of the voxel values, the aligned pairs of voxel values comprising a voxel value from the first image and a spatially corresponding voxel value from the second image, each mutation probability being related to the likelihood that a voxel value in the first image corresponds to a spatially corresponding voxel value in the second image and vice versa, said defining being based on a selected geometric relationship of the first and second images; ^{'},

(b) selecting a first transform defining a geometric relationship of the second image relative to the first image ;

(c) calculating (320, 344) a measure of the likelihood for a predetermined set of aligned voxel pairs using the mutation probabilities, the measure of the likelihood being an indicium of the probability of obtaining the first image given the second image and vice versa;

(d) selecting a different transform (332, 360) defining a geometric relationship of the second image relative to the first image; and

(e) iteratively repeating steps (c) and (d) until an optimal transform (150) defining a geometric relationship of the second image relative to the first image is calculated, the optimal transform providing an optimal measure of the likelihood.

2. The method of claim 1, further including at least one step selected from:

storing data representative of the optimal transform; and registering the first and second images using the optimal transform.

3. The method of either one of claims 2 and 3 , further including the step of :

displaying a composite image formed from the first and second images .

4. The method of any one of claims 1-3, wherein the optimal transform is selected from a rigid-body transform, an affine transform, a warping transform, and a nonlinear transform.

5. The method of any one of claims 1-4, wherein the step of defining mutation probabilities includes accessing stored pre-computed mutation probability values (140) .

6. The method of any one of claims 1-5, wherein the step of defining mutation probabilities includes estimating mutation probabilities (340) based on a currently selected transform defining a geometric relationship of the second image relative to the first image.

7. The method of any one of claims 1-6, wherein the plurality of aligned pairs comprises all aligned pairs in overlapping portions of the first and second images.

8. The method of any one of claims 1-7, wherein the plurality of pairs are limited to one or more selected subvolumes in overlapping portions of the first and second images .

9. The method of any one of claims 1-8, wherein the plurality of pairs are limited to aligned pairs in overlapping portions of the first and second images having voxel values within a preselected range of values .

10. The method of any one of claims 1-9, wherein the measure of the likelihood is symmetrical.

11. The method of any one of claims 1-10, wherein the step of calculating the measure of the likelihood includes summing logarithms of the mutation probabilities.

12. The method of any one of claims 1-11, wherein the step of calculating the measure of the likelihood includes:

calculating, as a sum of logarithms of the mutation probabilities, a first logarithmic likelihood value being an indicium of the probability of obtaining the first image given the second image for the predetermined set of aligned voxel pairs;

calculating, as a sum of logarithms of the mutation probabilities, a second logarithmic likelihood value being an indicium of the probability of obtaining the second image given the first image for the predetermined set of aligned voxel pairs; and

adding the first and second logarithmic likelihood values .

13. The method of claim 12, wherein adding the first and second logarithmic likelihood values produces a summed likelihood value within a preselected range, and further wherein the summed likelihood value increases within said range with increasing registration quality.

14. The method of any one of claims 1-13, further including:

normalizing the calculated measure of likelihood for a predetermined set of aligned voxel pairs using the mutation probabilities to produce a normalized measure of the likelihood, the normalized measure of the likelihood being an indicium of the probability of obtaining the first image given the second image and of obtaining the second image given the first image, the normalized measure being bounded within a preselected and finite range of values; and output ing the normalized measure, wherein the normalized measure approaches one end of the finite range as the quality of the registration increases, the normalized measure thereby serving to identify to the user the quality of the registration.

15. An image processing system (100) for registering a first volumetric image (110) and a second volumetric image

(112) , the volumetric images comprising three-dimensional arrays of voxel values, comprising:

a registration processor (130) and associated memory for storing a plurality of volumetric image representations to be registered, the registration processor:

determining mutation probabilities (316,

340) for a plurality of aligned pairs of the voxel

values, the aligned pairs of voxel values comprising

a voxel value from the first image and a spatially

corresponding voxel yalue from the second image, each

mutation probability being related to the likelihood

that a voxel value in the first image corresponds to

a spatially corresponding voxel value in the second

image and vice versa, said determining being based on

a selected geometric relationship of the first and

second images;

calculating a measure of the likelihood

(320, 344) for a plurality of geometric relationships

between the first and second images, the measure of

likelihood being calculated for a predetermined set

of aligned voxel _{(} pairs using the mutation

probabilities, and the measure of the likelihood

being an indicium of the probability of obtaining the

first image given the second image and vice versa;

and

optimizing the measure of likelihood (328,

352) to find an .optimal transform defining a geometric relationship between the first and second

images;

a memory (150) coupled to the registration processor for storing parameters representative of the optimal transform; and

a display system (160, 162) for forming a composite image representation of the first and second images.

16. The image processing system of claim 15, further comprising:

a diagnostic imaging scanner.

17. The image processing system of claim 16, wherein the diagnostic imaging scanner comprises an MR scanner, an x-ray CT scanner, and PET scanner, a SPECT scanner, an ultrasound scanner, or a combination thereof.

18. The image processing system of any one of claims

15-17, wherein the determining of mutation probabilities includes accessing stored pre-computed mutation probability values (140) .

, 19. The image processing system of any one of claims 15-18, wherein the defining of mutation probabilities includes estimating (340) mutation probabilities based on a currently selected transform that defines a geometric relationship of the second image relative to the first image.

20. The image processing system of any one of claims 15-19, wherein the measure of the likelihood is symmetrical.