Segmentation of tissue images using color and texture

• US 7,949,181 B2
• Filed: 06/28/2007
• Issued: 05/24/2011
• Est. Priority Date: 06/28/2007
• Status: Expired due to Fees

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Padfield, Dirk Ryan, Cline, Harvey, Datar, Manasi
• Primary Examiner(s): Ahmed; Samir A
• Assistant Examiner(s): Liu; Li
• Application Number:
  US11/770,323
• Publication Number:
  US 20090003691A1
• Time in Patent Office: 1,426 Days
• Field of Search: None
• US Class Current: 382/164
• CPC Class Codes: G06V 20/695 Preprocessing, e.g. image segmentation

Abstract

The present techniques provide for the processing of color tissue images based on image segmentation. In an exemplary embodiment, the color and texture features of pixels in a tissue image are used to generate a matrix of feature vectors. A subset of feature vectors is selected from the matrix of feature vectors and a set of colors and textures are derived using the tissue image and the subset of feature vectors. An initial segmented tissue image is then generated from this set of colors and textures.

Claims

1. A method for processing a tissue image of stained tissue, the method comprising:

generating a matrix of feature vectors, wherein each feature vector corresponds to a respective pixel of the tissue image and describes color and texture features of the respective pixel, wherein the stained tissue depicted in the tissue image is stained such that each color present in the stained tissue corresponds to a different tissue type;

selecting a subset of feature vectors from the matrix of feature vectors;

deriving a set of colors and textures using the tissue image and the subset of feature vectors; and

generating an initial segmented tissue image using the set of colors and textures and the tissue image; and

merging regions within the initial segmented tissue image to generate a final segmented image having a number of segments corresponding to the number of tissue types in the tissue image, wherein the number of segments is predetermined based on the number of different tissue types in the stained tissue.

2. The method as recited in claim 1, comprising extracting the color and texture features for each pixel of interest in the tissue image.

3. The method as recited in claim 1, wherein the color and texture features comprise three color features and four texture features.

4. The method as recited in claim 1, wherein deriving the set of colors and textures comprises performing a two-dimensional self organized mapping using the subset of feature vectors and the tissue image.

5. The method as recited in claim 1, wherein generating the initial segmented tissue image comprises performing a one-dimensional self organized mapping using the set of colors and textures and the tissue image.

6. The method as recited in claim 1, comprising assigning colors or labels to the segments of the final segmented image to generate a pseudo-colored or labeled image.

7. The method as recited in claim 1, wherein merging regions is iteratively performed until the predetermined number of segments remain.

8. The method as recited in claim 1, comprising performing an automatic pathology assessment of the initial segmented tissue image or the final segmented image.

9. The method as recited in claim 1, comprising establishing a training database using the initial segmented tissue image or the final segmented image.

10. The method as recited in claim 1, comprising training a pathologist or a CAD algorithm using the initial segmented tissue image or the final segmented image.

11. One or more non-transitory computer-readable media having a computer program comprising:

a routine configured to generate a matrix of feature vectors, wherein each feature vector corresponds to a respective pixel of a tissue image and describes color and texture features of the respective pixel, wherein the stained tissue depicted in the tissue image is stained such that each color present in the stained tissue corresponds to a different tissue type;

a routine configured to select a subset of feature vectors from the matrix of feature vectors;

a routine configured to derive a set of colors and textures using the tissue image and the subset of feature vectors; and

a routine configured to generate an initial segmented tissue image using the set of colors and textures and the tissue image; and

a routine configured to merge regions within the initial segmented tissue image to generate a final segmented image having a number of segments corresponding to the number of tissue types in the tissue image, wherein the number of segments is predetermined based on the number of different tissue types in the stained tissue.

12. The one or more non-transitory computer-readable media as recited in claim 11, wherein the computer program comprises a routine configured to extract the color and texture features for each pixel of interest in the tissue image.

13. The one or more non-transitory computer-readable media as recited in claim 11, wherein the routine configured to derive the set of colors and textures performs a two-dimensional self organized mapping using the subset of feature vectors and the tissue image.

14. The one or more non-transitory computer-readable media as recited in claim 11, wherein the routine configured to generate an initial segmented tissue image performs a one-dimensional self organized mapping using the set of colors and textures and the tissue image.

15. The one or more non-transitory computer-readable media as recited in claim 11, wherein the computer program comprises a routine configured to assign colors or labels to the segments of the final segmented image to generate a pseudo-colored or labeled image.

16. The one or more non-transitory computer-readable media as recited in claim 11, wherein the routine configured to merge regions iteratively merges regions until the predetermined number of segments remain.

17. An image analysis system, comprising:

an imager configured to optically analyze a stained tissue sample wherein the stained tissue sample is stained such that each color present in the sample corresponds to a different tissue type;

data acquisition circuitry configured to acquire a tissue image of the stained tissue sample via the imager; and

data processing circuitry configured to generate a matrix of feature vectors, wherein each feature vector corresponds to a respective pixel of the tissue image and describes color and texture features of the respective pixel, to select a subset of feature vectors from the matrix of feature vectors, to derive a set of colors and textures using the tissue image and the subset of feature vectors, to generate an initial segmented tissue image using the set of colors and textures and the tissue image; and

to merge regions within the initial segmented tissue image to generate a final segmented image having a number of segments corresponding to the number of tissue types in the tissue image, wherein the number of segments is predetermined based on the number of different tissue types in the stained tissue.

18. The image analysis system of claim 17, wherein the data processing circuitry is further configured to extract the color and texture features for each pixel of interest in the tissue image.

19. The image analysis system of claim 17, wherein the data processing circuitry is configured to derive the set of colors and textures by performing a two-dimensional self organized mapping using the subset of feature vectors and the tissue image.

20. The image analysis system of claim 17, wherein the data processing circuitry is configured to generate the initial segmented tissue image by performing a one-dimensional self organized mapping using the set of colors and textures and the tissue image.

21. The image analysis system of claim 17, wherein the data processing circuitry is configured to assign colors or labels to the segments of the final segmented image to generate a pseudo-colored or labeled image.

22. The image analysis system of claim 17, wherein the data processing circuitry is configured to merge regions iteratively until the predetermined number of segments remain.