System and method for image registration using nonparametric priors and statistical learning techniques

Abstract

A method for image registration includes receiving first and second image information. A library of joint intensity distributions, spanning a space of non-parametric statistical priors, derived from earlier perfect matching results is received. From among this library, a preferred learned joint intensity distribution is automatically selected during the registration process. As a result, a displacement field is generated both (i) maximizing the statistical dependency between an intensity distribution of the first and second image information and (ii) minimizing the statistical distance to the learned joint intensity distributions. The generated displacement field is used to transform an image structure from the first image information to an image structure of the second image information.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

A more complete appreciation of the present disclosure and many of the attendant advantages thereof will be readily obtained as the same becomes better understood by reference to the following detailed description when considered in connection with the accompanying drawings, wherein:

FIG. 1 shows a schematic plot of energy according to an exemplary embodiment of the present invention;

FIG. 2 shows the registration performance analysis of the SPECT slice and the CT image;

FIGS. 3A-3D illustrate different registration problems characterized by different joint intensity distributions;

FIGS. 4A-4C illustrate the use of several prior distributions according to an exemplary embodiment of the present invention;

FIGS. 5A-5H illustrate facial images used for training and registration;

FIGS. 6A-6F illustrate facial image registration results; and

FIG. 7 shows an example of a computer system capable of implementing the method and apparatus according to embodiments of the present disclosure.

Claims

1. A method for image registration, comprising: receiving first image information;receiving second image information;automatically selecting a preferred learned joint intensity distribution from among a library of learned joint intensity distributions;generating a displacement field both (i) maximizing the statistical dependency between an intensity distribution of the first and second image information and (ii) minimizing the distance to a learned joint intensity distribution; andusing the generated displacement field to transform an image structure from the first image information to an image structure of the second image information.

2. The method of claim 1, wherein a library of joint intensity distributions spans a space of non-parametric statistical priors derived from earlier registrations.

3. The method of claim 1, wherein the first image information represents a first medical image and the second medical information represents a second medical image.

4. The method of claim 3, wherein the first medical image is an image of a subject taken with a first medical imaging device and the second medical image is an image of the subject taken with a second medical imaging device different than the first medical imaging device.

5. The method of claim 3, wherein the first medical image is an image of a subject taken at a first time and the second medical image is an image of the subject taken at a second time later than the first point in time.

6. The method of claim 1, further comprising adding a joint intensity distribution to the library of learned joint intensity distributions based on the generated displacement field.

7. The method of claim 1, wherein the ability to accurately generating a displacement field increases as the library of learned joint intensity distributions increases.

8. The method of claim 1, wherein the automatically selecting a preferred learned joint intensity distribution from among a library of learned joint intensity distributions comprises minimizing the energy E for the displacement field u, according to the formula: E(u)=Eprior(u)+α1EMI(u)+α2Esmooth(u), wherein

9. A system for image recognition, comprising: receiving first image information;receiving second image information;automatically selecting a preferred learned joint intensity distribution from among a library of learned joint intensity distributions;generating a displacement field both (i) maximizing the statistical dependency between an intensity distribution of the first and second image information and (ii) minimizing the distance to a learned joint intensity distribution; andusing the generated displacement field to transform an image structure from the first image information to an image structure of the second image information.

10. The system of claim 9, wherein a library of joint intensity distributions spans a space of non-parametric statistical priors derived from earlier registrations.

11. The system of claim 9, wherein the first image information represents a first medical image and the second medical information represents a second medical image.

12. The system of claim 11, wherein the first medical image is an image of a subject taken with a first medical imaging device and the second medical image is an image of the subject taken with a second medical imaging device different than the first medical imaging device.

13. The system of claim 11, wherein the first medical image is an image of a subject taken at a first time and the second medical image is an image of the subject taken at a second time later than the first point in time.

14. The system of claim 9, further comprising an adding unit to add a joint intensity distribution to the library of learned joint intensity distributions based on the generated displacement field.

15. The system of claim 9, wherein the ability to accurately generating a displacement field increases as the library of learned joint intensity distributions increases.

16. The system of claim 9, wherein the selecting unit automatically selects a preferred learned joint intensity distribution from among a library of learned joint intensity distributions comprises minimizing the energy E for the displacement field u, according to the formula: E(u)=Eprior(u)+α1EMI(u)+α2Esmooth(u), wherein

17. A computer system comprising: a processor; anda program storage device readable by the computer system, embodying a program of instructions executable by the processor to perform method steps for image registration, the method comprising:receiving first image information;receiving second image information;automatically selecting a preferred learned joint intensity distribution from among a library of learned joint intensity distributions;generating a displacement field both (i) maximizing the statistical dependency between an intensity distribution of the first and second image information and (ii) minimizing the distance to a learned joint intensity distribution; andusing the generated displacement field to transform an image structure from the first image information to an image structure of the second image information.

18. The computer system of claim 17, wherein the selected preferred joint intensity distribution is used as an initial estimate of the intensity distribution of the first image information and the second image information.

19. The computer system of claim 17, wherein a library of joint intensity distributions spans a space of non-parametric statistical priors derived from earlier registrations.

20. The computer system of claim 19, wherein the first medical image is an image of a subject taken with a first medical imaging device and the second medical image is an image of the subject taken with a second medical imaging device different than the first medical imaging device.

21. The computer system of claim 19, wherein the first medical image is an image of a subject taken at a first time and the second medical image is an image of the subject taken at a second time later than the first point in time.

22. The computer system of claim 17, further comprising adding a joint intensity distribution to the library of learned joint intensity distributions based on the generated displacement field.

23. The computer system of claim 17, wherein the ability to accurately generating a displacement field increases as the library of learned joint intensity distributions increases.

24. The computer system of claim 17, wherein the automatically selecting a preferred learned joint intensity distribution from among a library of learned joint intensity distributions comprises minimizing the energy E for the displacement field u, according to the formula: E(u)=Eprior(u)+α1EMI(u)+α2Esmooth(u), wherein