System and method for boundary classification and automatic polyp detection

Patent metadata

Field	Value
Publication number	US-9741116-B2
Application number	US-201414914896-A
Country	US
Kind code	B2
Filing date	Aug 28, 2014
Priority date	Aug 29, 2013
Publication date	Aug 22, 2017
Grant date	Aug 22, 2017

Abstract

A system and method is provided for automated polyp detection in optical colonoscopy images. The system includes an input configured to acquire a series of optical images, and a processor configured to process the optical images. Processing steps include performing a boundary classification with steps comprising locating a series of edge pixels using at least one acquired optical image, selecting an image patch around each said edge pixel, performing a classification threshold analysis on each image patch of said edge pixels using a set of determined boundary classifiers, and identifying, based on the classification threshold analysis, polyp edge pixels consistent with a polyp edge. Processing steps for the processor also include performing a vote accumulation, using the identified polyp edge pixels, to determine a polyp location. The system also includes an output configured to indicate potential polyps using the determined polyp location.

First claim

We claim:

1. A system for automated polyp detection in optical colonoscopy images comprising:
an input configured to acquire multiple optical images;
a processor configured to process said optical images with steps comprising:
i. performing a boundary classification with steps comprising:
a. locating a series of edge pixels using at least one acquired optical image;
b. selecting an image patch around each said edge pixel;
c. performing a classification threshold analysis on each image patch of said edge pixels using a set of determined boundary classifiers;
d. identifying, based on the classification threshold analysis, polyp edge pixels consistent with a polyp edge;

ii. performing a vote accumulation using the identified polyp edge pixels to determine a polyp location;
an output configured to indicate potential polyps using the determined polyp location;
wherein performing the classification threshold analysis of said image patches comprises a comparison between said image patches and a series of image patterns; and
wherein each of the series of image patterns comprises a plurality of categories, with each category spanning an angular orientation consistent with ⅙ of a range between 0 and π.

2. The system of claim 1 , wherein the processor is further configured to apply a color filter to create a plurality of color filtered images for each optical image.

3. The system of claim 2 , wherein said color filter comprises a red filter, a green filter, and a blue filter.

4. The system of claim 1 , wherein locating said edge pixels comprises applying a Canny edge detection algorithm.

5. The system of claim 1 , wherein the processor is further configured to estimate an edge direction for each of the series of said edge pixels.

6. The system of claim 5 , wherein each edge direction is estimated by a tensor voting with a ball tensor placed at a location of each said edge pixel.

7. The system of claim 1 , wherein each image patch comprises a plurality of pixels arranged such that the said edge pixel is positioned central to the image patch.

8. The system of claim 1 , wherein identifying polyp edge pixels consistent with a polyp edge comprises applying a random forest classifier.

9. The system of claim 1 , wherein the processor is further configured to determine a classification confidence for the identified polyp edge pixels.

10. The system of claim 1 , wherein for each edge pixel in the optical image, said vote accumulation is performed for a plurality of categories, generating a plurality of voting maps.

11. The system of claim 10 , wherein the plurality of voting maps are combined according to:
argmaxΠ i N Σ vεV i M v

where N is a number of categories, M v is a vote accumulation intensity, and v is a voter in a voting category, V i .

12. The system of claim 1 , wherein the processor is further configured to perform a ray back-projection technique to indicate a probability for a true detection of the potential polyps.

13. A method for automated polyp detection in optical colonoscopy images comprising:
i. performing a boundary classification with steps comprising:
a. locating a series of edge pixels using at least one acquired optical image;
b. selecting an image patch around each said edge pixel;
c. performing a classification threshold analysis on each image patch of said edge pixels using a set of determined boundary classifiers;
d. identifying, based on the classification threshold analysis, polyp edge pixels consistent with a polyp edge;

ii. performing a vote accumulation using the identified polyp edge pixels to determine a polyp location;
iii. generating a report indicative of potential polyps using the determined polyp location
wherein performing the classification threshold analysis of said image patches comprises a comparison between said image patches and a series of image patterns; and
wherein each of the series of image patterns comprises a plurality of categories, with each category spanning an angular orientation consistent with ⅙ of a range between 0 and π.

14. The method of claim 13 , the method further comprising applying a color filter to create a plurality of color filtered images for each optical image.

15. The method of claim 14 , wherein said color filter comprises a red filter, a green filter, and a blue filter.

16. The method of claim 13 , wherein locating said edge pixels comprises applying a Canny edge detection algorithm.

17. The method of claim 13 , the method further comprising estimating an edge direction for each of the series of said edge pixels.

18. The method of claim 17 , wherein each edge direction is estimated by a tensor voting with a ball tensor placed at a location of each said edge pixel.

19. The method of claim 13 , wherein each image patch comprises a plurality of pixels arranged such that the said edge pixel is positioned central to the image patch.

20. The method of claim 13 , wherein identifying polyp edge pixels consistent with a polyp edge comprises applying a random forest classifier.

21. The method of claim 13 , the method further comprising determining a classification confidence for the identified polyp edge pixels.

22. The method of claim 13 , wherein for each edge pixel in the optical image, said vote accumulation is performed for a plurality of categories, generating a plurality of voting maps.

23. The method of claim 22 , wherein the plurality of voting maps are combined according to:
argmaxΠ i N Σ vεV i M v

where N is a number of categories, M v is a vote accumulation intensity, and v is a voter in a voting category, V i .

24. The method of claim 13 , the method further comprising performing a ray back-projection technique to indicate a probability for a true detection of the potential polyps.

Assignees

• Liang Jianming
• Tajbakhsh Nima
• Gurudu Suryakanth R
• Mayo Foundation
• Univ Arizona State

Inventors

• Liang Jianming
• Tajbakhsh Nima
• Gurudu Suryakanth R

Classifications

• A61B1/000094 Primary

  extracting biological structures · CPC title

• G06T7/0016 Primary

  involving temporal comparison · CPC title

• G06T5/20

  using local operators · CPC title

• A61B1/31

  for the rectum, e.g. proctoscopes, sigmoidoscopes {, colonoscopes} · CPC title

• G06T7/12

  Edge-based segmentation · CPC title