Method and apparatus for stain separation using vector analysis

• US 9,036,889 B2
• Filed: 07/13/2012
• Issued: 05/19/2015
• Est. Priority Date: 07/13/2012
• Status: Active Grant

• Current Assignee: Sony Corporation (Sony Group Corp.)
• Original Assignee: Sony Corporation (Sony Group Corp.)
• Inventors: Liu, Ming-Chang, Zhang, Ximin
• Primary Examiner(s): Drennan, Barry
• Assistant Examiner(s): BEG, SAMAH A
• Application Number:
  US13/549,019
• Publication Number:
  US 20140016853A1
• Time in Patent Office: 1,040 Days
• Field of Search: None
• US Class Current: 382/133
• CPC Class Codes: G06T 2207/10056 Microscopic image
• CPC Class Codes: G06T 2207/20048 Transform domain processing
• CPC Class Codes: G06T 2207/30024 Cell structures in vitro; Tissue sections in vitro
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/168 involving transform domain methods
• CPC Class Codes: G06T 7/90 Determination of colour characteristics

Abstract

A computer-implemented method and apparatus for stain separation of a pathology image using stain vector analysis comprising converting an original image into an optical domain image, performing stain vector analysis on the optical domain image to obtain one or more stain vectors, deconvoluting the vectors adaptively to produce one or more separated stain images.

Claims

1. A computer-implemented method for stain separation of a pathology image using stain vector analysis comprising:

converting an original image into an optical domain image;

performing stain vector analysis on the optical domain image to obtain one or more stain vectors;

deconvoluting the one or more stain vectors to produce one or more separated stain images;

isolating stain amounts in each color channel of the one or more separated stain images;

comparing, for each pixel in the original image, a pixel value from one of the one or more stain images to a pixel value in another image of the one or more stain images; and

selecting, for each pixel in the original image, a pixel value with a largest stain value, wherein a plurality of selected pixel values forms a strongest stain image.

2. The method of claim 1 wherein performing stain vector analysis comprises:

calculating an image histogram based on the optical domain image;

calculating a first and second color index from the image histogram, where an optical domain value of the histogram is greater than a predefined threshold;

determining a first stain vector based on a first and second vector formed by the first and second color index;

estimating a second stain vector by selecting a color index forming a first largest angle with the first stain vector; and

estimating a third stain vector by selecting a color index forming a second largest angle with a plane formed by the first and second stain vector.

3. The method of claim 2 wherein the image histogram is determined according to the following formula:

4. The method of claim 2 wherein the first color index and second color index are calculated according to:

5. The method of claim 1 further comprising selecting a region of the optical domain image to perform the stain vector analysis by:

scaling the optical domain image from an original size to a different size;

selecting a window of a predetermined size of the optical domain image which is a most salient window as compared to all other windows of the predetermined size of the optical domain image; and

projecting the window back to the original size to obtain the region.

6. The method of claim 5 wherein the most salient window is obtained by finding a minimum sum of red and blue channels.

7. The method of claim 1 further comprising a user'"'"'s manual selection of a region of the optical domain image upon which to perform the stain vector analysis.

8. The method of claim 1 wherein the deconvoluting comprises:

determining one or more stain amounts by multiplying the optical domain image with the one or more stain vectors.

9. The method of claim 1 further comprising:

performing adaptive separation on the one or more stain images wherein each pixel position of the original image is associated with only one of the one or more stain images.

10. The method of claim 9 wherein the adaptive separation further comprises:

determining a first, second and third stain amount from a first, second and third stain image among the one or more stain images; and

selecting, as a stain amount for a final image, the largest of the first, second or third stain amount and resetting a pixel value for the stain images corresponding to unselected stain amounts to a max color value.

11. An apparatus for stain separation of a pathology image using stain vector analysis, the apparatus comprising a processor that executes:

a transform module that converts an original image into an optical domain image;

a vector module, coupled to the transform module, that performs stain vector analysis on the optical domain image to obtain one or more stain vectors;

a deconvolution module, coupled to the vector module, that deconvolutes the one or more vectors to produce one or more separated stain images; and

an adaptive separation module that isolates stain amounts in each color channel of the one or more separated stain images and for each pixel in the original image, compares, for each pixel in the original image, a pixel value from one of the one or more stain images to a pixel value in another of the one or more stain images; and

selects a pixel value from one of the one or more stain images with a largest stain value, wherein a plurality of selected pixel values form a strongest stain image.

12. The apparatus of claim 11, wherein the processor further executes:

a histogram module, coupled to the transform module, that calculates an image histogram based on the optical domain image;

a histogram color indexing module, coupled to the vector module, that calculates a first and second color index from the image histogram, where an optical domain value of the histogram is greater than a predefined threshold;

a vector determination module, coupled to the histogram color indexing module, that;

determines a first stain vector based on a first and second vector formed by the first and second color index;

estimates a second stain vector by selecting a color index forming a largest angle with the first stain vector; and

estimates a third stain vector by selecting a color index forming the largest angle with a plane formed by the first and second stain vector.

13. The apparatus of claim 12 wherein the image histogram is determined according to the following formula:

14. The apparatus of claim 12 wherein the first color index and the second color index are calculated according to:

15. The apparatus of claim 11, wherein the processor further executes a selection module that selects a region of the optical domain image to perform the stain vector analysis, comprising:

a scaling module that scales the optical domain image from an original size to a different size;

a color module, coupled to the scaling module, that selects a window of a predetermined size of the optical domain image with a most salient red and blue channel color compared to all other windows of the predetermined size of the optical domain image; and

the scaling module further projects the window back to the original size to obtain the region.

16. The apparatus of claim 15 wherein the most salient red and blue channel color are obtained by finding a minimum of a summation in both red and blue channels.

17. The apparatus of claim 11 wherein the deconvolution module further determines one or more stain amounts by multiplying the optical domain image with the one or more stain vectors.

18. The apparatus of claim 11 wherein the adaptive separation module, coupled to the deconvolution module, performs adaptive separation on the one or more stain images wherein each pixel position of the original image is associated with only one of the one or more stain images.

19. The apparatus of claim 18 wherein the adaptive separation module further:

determines a first, second and third stain amount from a first, second and third stain image among the one or more stain images; and

selects, as a stain amount for a final image, the largest of the first, second or third stain amount and resets a pixel value for the stain images corresponding to unselected stain amounts to a max color value.

20. The apparatus of claim 11, wherein the processor further executes a user selection module, coupled to the transform module, that receives a user'"'"'s manual selection of an area of the optical domain image for input to the vector module.