Method for determining a property map of an object, particularly of a living being, based on at least a first image, particularly a magnetic resonance image

• US 8,290,568 B2
• Filed: 01/09/2009
• Issued: 10/16/2012
• Est. Priority Date: 07/12/2006
• Status: Active Grant

• Current Assignee: Eberhard-Karls-Universitat Tubingen Universitatsklinikum, Max-Planck-Gesellschaft Zur Foerderung Der Wissenschaften e.V. (Max-Planck-Gesellschaft zur Frderung der Wissenschaften e)
• Original Assignee: Eberhard-Karls-Universitat Tubingen Universitatsklinikum
• Inventors: Pichler, Bernd, Steinke, Florian, Hofmann, Matthias, Schölkopf, Bernhard
• Primary Examiner(s): Le, Long V.
• Assistant Examiner(s): IMPINK, BRADLEY GERALD
• Application Number:
  US12/351,429
• Publication Number:
  US 20100049032A1
• Time in Patent Office: 1,376 Days
• Field of Search: None
• US Class Current: 600/411
• CPC Class Codes: A61B 6/037 Emission tomography
• CPC Class Codes: G01R 33/481 MR combined with positron emission tomography [PET] or single photon emission computed tomography [SPECT]
• CPC Class Codes: G06T 2207/10088 Magnetic resonance imaging [MRI]
• CPC Class Codes: G06T 2207/30004 Biomedical image processing
• CPC Class Codes: G06T 2207/30196 Human being; Person
• CPC Class Codes: G06T 7/33 using feature-based methods

Abstract

It is disclosed a system and method (12) for determining a property map (82) of an object, particularly a human being, based on at least a first image (84), particularly an magnetic resonance (MR) image, of the object. In the method (12), a structure of reference pairs is defined in a first step (96), wherein each reference pair (16-26) comprises at least two entries (62). The first entry represents a property value, particularly an attenuation value. The second entry (62) preferably represents a group of image points (67) belonging together, which is extracted particularly from MR images (28) and comprises an interesting image point corresponding to the property value. In another step (98) of the method (12) a plurality of training pairs (16-26) is provided. A structure of the training pairs (16-26) corresponds to the structure of reference pairs, and the entries of respective training pairs (16-26) are known. In another step (100) of the method (12), an assignment between the first entries and the other entries (62-66) of the training pairs (16-26) is determined by machine learning, thus allowing prediction of a property value (88) corresponding to an arbitrary point (90) of the first image (84).

Claims

1. A method for determining a property map of an object based on at least a first magnetic resonance (MR) image of the object, comprising:

defining a structure of reference pairs, wherein each reference pair has at least first and second entries, wherein each of said first entries represents an unknown property value, and wherein each of said second entries represents a group of image points belonging together, which are extracted from said first MR image recorded by a scanner, wherein the group of image points includes at least one interesting image point which corresponds to one of the unknown property-values, wherein each unknown property value represents an attenuation value;

providing a plurality of training pairs, wherein a structure of the training pairs corresponds to the structure of said reference pairs, and wherein first and second entries of the respective training pairs are known; and

determining an assignment between the first entries and the second entries of the training pairs by machine learning, and predicting an attenuation value for an arbitrary image point which corresponds to the at least one interesting image point of the first MR image by applying the determined assignment to one of the second entries of the reference pairs including said arbitrary image point.

2. The method of claim 1, wherein the structure of said reference pairs additionally comprises third or additional entries, which are taken into account during the step of determining the assignment.

3. The method of claim 2, wherein the third or additional entries are selected from a group consisting of:

a registered coordinate;

another group of image points belonging together of at least a second image, wherein the other group of image points of the at least second image shows at least partially an identical object patch like the group of image points of the first image; and

a class affiliation.

4. The method of claim 3, wherein the registered coordinate is defined by one of a coordinate of a point equivalent to the interesting point in a reference object, and one or more distances of a point equivalent to the interesting point relative to anatomically characteristic positions within the reference object.

5. The method of claim 3, wherein a similarity between the registered coordinate and a position-dependent distance measure is determined, which is dependent on properties of a reference object as well as a position of an equivalent point within the reference object.

6. The method of claim 3, wherein the at least second image is one of a positron emission tomography (PET) image and another MR image recorded by a scanner.

7. The method of claim 3, wherein at least one of the at least first MR image and the at least second image are pre-processed by the extraction of features.

8. The method of claim 7, wherein the pre-processing is performed a Fourier/Wavelet transformation, SIFT, edge detectors, or contrast amplification.

9. The method of claim 3, wherein additionally at least the second image is used for registering with the at least first image, and for extending the at least first image.

10. The method of claim 1, wherein said second entries represent registered coordinates.

11. The method the claim 1, wherein the group of image points belonging together defines a window.

12. The method of claim 11, wherein the window is rectangular.

13. The method of claim 1, wherein a subgroup is selected from the group of image points belonging together, which comprises the interesting image point.

14. The method of claim 1, wherein each image point has an intensity value.

15. The method of claim 14, wherein an intensity histogram is established based on a subgroup, which can be added to the structure as another entry.

16. The method of claim 1, wherein a texture is determined for each group of image points belonging together.

17. The method of claim 1, wherein an intensity average is formed based on intensity values of at least the image points of the group of image points belonging together, the intensity average being added to the structure together with the intensity of the interesting image points as another entry.

18. A system for determining a property map of an object based on at least a first magnetic resonance (MR) image of the object, comprising:

an imaging system;

a computer system connected to the imaging system;

wherein the computer system is configured to execute a series of steps to determine the property map of the object, the steps comprising;

defining a structure of reference pairs, wherein each reference pair has at least first and second entries, wherein each of said first entries represents an unknown property value, and wherein each of said second entries represents a group of image points belonging together, which are extracted from said first MR image recorded by a scanner, wherein the group of image points includes at least one interesting image point which corresponds to one of the unknown property values, wherein each unknown property value represents an attenuation value;

providing a plurality of training pairs, wherein a structure of the training pairs corresponds to the structure of said reference pairs, and wherein first and second entries of the respective training pairs are known; and

determining an assignment between the first entries and the second entries of the training pairs by machine learning, and predicting an attenuation value for an arbitrary image point which corresponds to the at least one interesting image point of the first MR image by applying the determined assignment to one of the second entries of the reference pairs including said arbitrary image point.

19. A non-transitory computer readable medium comprising a set of instructions that when executed by a computer processor, result in a series of steps being performed to determine a property map of an object based on at least a first magnetic resonance (MR) image of the object, the steps comprising:

defining a structure of reference pairs, wherein each reference pair has at least first and second entries, wherein each of said first entries represents an unknown property value, and wherein each of said second entries represents a group of image points belonging together, which are extracted from said first MR image recorded by a scanner, wherein the group of image points includes at least one interesting image point which corresponds to one of the unknown property values, wherein each unknown property value represents an attenuation value;

providing a plurality of training pairs, wherein a structure of the training pairs corresponds to the structure of said reference pairs, and wherein first and second entries of the respective training pairs are known; and

determining an assignment between the first entries and the second entries of the training pairs by machine learning, and predicting an attenuation value for an arbitrary image point which corresponds to the at least one interesting image point of the first MR image by applying the determined assignment to one of the second entries of the reference pairs including said arbitrary image point.