SYSTEM AND METHOD FOR AUTOMATED DISEASE ASSESSMENT IN CAPSULE ENDOSCOPY

• US 20120316421A1
• Filed: 07/07/2010
• Published: 12/13/2012
• Est. Priority Date: 07/07/2009
• Status: Abandoned Application

• Current Assignee: Johns Hopkins University
• Original Assignee: Johns Hopkins University
• Inventors: Kumar, Rajesh, Hager, Gregory, Dassopoulos, Themistocles, Girgis, Hani
• Application Number:
  US13/382,855
• Publication Number:
  US 20120316421A1
• Time in Patent Office: Days
• US Class Current: 600/407
• CPC Class Codes: A61B 1/000094 extracting biological structures
• CPC Class Codes: A61B 1/000096 using artificial intelligence
• CPC Class Codes: A61B 1/041 Capsule endoscopes for imaging
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/10068 Endoscopic image
• CPC Class Codes: G06T 2207/30032 Colon polyp
• CPC Class Codes: G06T 7/0012 Biomedical image inspection

Abstract

A system and method for automated image analysis which may enhance, for example, capsule endoscopy diagnosis. The system and methods may reduce the time required for diagnosis, and also help improve diagnostic consistency using an interactive feedback tool. Furthermore, the system and methods may be applicable to any procedure where efficient and accurate visual assessment of a large set of images is required.

Claims

1. An automated method of processing images from an endoscope comprising:

receiving a plurality of endoscopic images by an image processing system;

processing each of said plurality of endoscopic images with said image processing system to determine whether at least one attribute of interest is present in each image that satisfies a predetermined criterion; and

classifying said plurality of endoscopic images into a reduced set of images each of which contains at least one attribute of interest and a remainder set of images each of which is free from said attribute.

2. The automated method according to claim 1, where the attribute of interest is a localized region of interest containing a disease relevant visual attribute.

3. The automated method of claim 2, wherein said disease relevant visual attribute comprises an image of:

a lesion, a polyp, bleeding, inflammation, discoloration, or stenosis.

4. The automated method according to claim 1, further comprising:

processing said reduced set of images with said image processing system to identify an attribute of interest in a first image of said reduced set of images that corresponds to an attribute of interest of a second image of said reduced set of images.

5. The automated method according to claim 4, further comprising:

classifying said reduced set of images into a non-redundant set of images such that no attribute of interest of any one of said non-redundant set of images corresponds to an attribute of interest of any other one of said non-redundant set of images.

6. The method according to claim 1, further comprising:

displaying result data with said image processing system, wherein said result data comprises an image from said reduced set of images containing at least one attribute of interest.

7. The method according to claim 6, further comprising:

receiving relevance feedback on said image processing system from an observer of said result data, wherein said relevance feedback comprises a change to said result data; and

training said image processing system based on said received relevance feedback.

8. The method according to claim 7, wherein said relevance feedback includes one or more of the following:

a change in said classification,

a removal of the image from said reduced set of images,

a change in an ordering of said reduced set of images,

an assignment of an assessment attribute, and

an assignment of a measurement.

9. The method according to claim 7, wherein said training comprises using at least one of the following:

artificial neural networks,

support vector machines, and

linear discriminant analysis.

10. The method according to claim 1, wherein said attribute of interest corresponds to an abnormality, said method further comprising:

assessing a severity of each said attribute of interest in said reduced set of images containing at least one attribute of interest using said image processing system.

11. The method according to claim 10, where said assessing comprises calculating one of:

a score,

a rank,

a structured assessment comprising of one or more categories,

a structured assessment on a Likert scale, and

a relationship with one or more other images, wherein said relationship comprises less severe or more severe.

12. The method according to claim 10, further comprising:

deriving a score for said reduced set of images containing at least one attribute of interest based on said severity of each said region of interest using said image processing system.

13. The method according to claim 12, wherein said score comprises at least one of:

a Lewis score,

a Crohn'"'"'s Disease Endoscopy Index of Severity,

a Simple Endoscopic Score for Crohn'"'"'s Disease,

a Crohn'"'"'s Disease Activity index, and

a rubric based on image appearance attributes, wherein said appearance attributes comprises one of;

lesion exudates, inflammation, color, and texture.

14. The method according to claim 1, further comprising:

prior to the first said processing, processing each of said plurality of endoscopic images with said image processing system to determine whether any of said plurality of endoscopic images is unusable for further processing; and

removing said unusable image from further processing.

15. The method according to claim 14, wherein said unusable image comprises at least one image of:

air bubbles,

food,

fecal matter,

normal tissue,

non-lesion, and

structures.

16. The method according to claim 1, wherein said processing each of said plurality of endoscopic images and classifying said plurality of endoscopic images comprises at least one of:

supervised machine learning and unsupervised machine learning.

17. The method according to claim 1, wherein said processing each of said plurality of endoscopic images comprises using at least one of:

statistical measures,

machine learning algorithms,

traditional classification techniques,

regression techniques,

feature vectors,

localized descriptors,

MPEG-7 visual descriptors,

edge features,

color histograms,

image statistics,

gradient statistics,

Haralick texture features,

dominant color descriptors,

edge histogram descriptors,

homogeneous texture descriptors,

spatial kernel weighting,

uniform grid sampling,

grid sampling with multiple scales,

local mode-seeking using mean shift,

generic lesion templates,

linear discriminate analysis,

logistic regression,

K-nearest neighbors,

relevance vector machines,

expectation maximation,

discrete wavelets, and

Gabor filters.

18. The method according to claim 1, wherein said predetermined criterion comprises a measurement of at least one of:

color,

texture,

hue,

saturation,

intensity,

energy,

entropy,

maximum probability,

contrast,

inverse difference moment, and

correlation.

19. The method according to claim 1, wherein said classifying said plurality of endoscopic images comprises using at least one of:

meta methods,

boosting methods,

bagging methods,

voting,

weighted voting,

adaboost,

temporal consistency,

performing a second classification procedure on data neighboring said localized region of interest, and

Bayesian analysis.

20. The method according to claim 1, wherein said endoscope comprises at least one of:

a wireless capsule endoscopy device,

an endoscope,

a flexible endoscope,

a contact hysteroscope,

a flexible borescope,

a video borescope,

a rigid borescope,

a pipe borescope,

a GRIN lens endoscope, and

a fibroscope.

21. The method according to claim 1, wherein,

said plurality of endoscopic images are images taken within a gastrointestinal track;

and

said attribute of interest comprises an anatomic abnormality in said gastrointestinal track.

22. The method according to claim 21, wherein said anatomic abnormality comprises at least one of:

a lesion,

mucosal inflammation,

an erosion,

an ulcer,

submucosal inflammation,

a stricture,

a fistulae,

a perforation,

an erythema,

edema,

blood, and

a boundary organ.

23. The method according to claim 1, wherein said receiving a plurality of endoscopic images by an image processing system comprises receiving said plurality of endoscopic images from one of:

a database of images, and

in real-time from said endoscope.

24. An endoscopy system, comprising:

an endoscope;

a processing unit in communication with said endoscope, said processing unit comprising executable instructions for detecting an attribute of interest;

wherein said processing unit performs the following in response to receiving a plurality of endoscopic images from said endoscope based on said executable instructions;

a determination of whether at least one attribute of interest is present in each image that satisfies a predetermined criterion; and

a classification of said plurality of endoscopic images into a reduced set of images each of which contains said at least one attribute of interest and a remainder set of images each of which is free from said at least one attribute of interest.

25. The system of claim 24, where the attribute of interest is a localized region of interest containing a disease relevant visual attribute.

26. The system of claim 25, wherein said disease relevant visual attribute comprises an image of:

a lesion, a polyp, bleeding, inflammation, discoloration, or stenosis.

27. The system of claim 24, wherein said processing unit further performs the following in response to receiving a plurality of endoscopic images from said endoscope based on said executable instructions:

an identification of an attribute of interest in a first image of said reduced set of images that corresponds to an attribute of interest of a second image of said reduced set of images.

28. The system of claim 27, wherein said processing unit further performs the following in response to receiving a plurality of endoscopic images from said endoscope based on said executable instructions:

a classification of said reduced set of images into a non-redundant set of images such that no attribute of interest of any one of said non-redundant set of images corresponds to an attribute of interest of any other one of said non-redundant set of images.

29. The system of claim 24, further comprising:

a display device; and

wherein said processing unit further performs the following in response to receiving a plurality of endoscopic images from said endoscope based on said executable instructions;

a display of result data on said display device, wherein said result data comprises an image from said reduced set of images containing at least one attribute of interest.

30. The system of claim 29, further comprising:

an input device; and

wherein said processing unit further performs the following in response to receiving a plurality of endoscopic images from said endoscope based on said executable instructions;

a receipt of relevance feedback, wherein said relevance feedback comprises a change to said result data; and

a training of said processing unit based on said received relevance feedback.

31. The system of claim 30, wherein said relevance feedback includes one or more of the following:

a change in said classification,

a removal of the image from said reduced set of images,

a change in an ordering of said reduced set of images,

an assignment of an assessment attribute, and

an assignment of a measurement.

32. The system of claim 30, wherein said training of said processing unit comprises using at least one of the following:

artificial neural networks,

support vector machines, and

linear discriminant analysis.

33. The system of claim 24, wherein

said attribute of interest corresponds to an abnormality;

and

wherein said processing unit further performs the following in response to receiving a plurality of endoscopic images from said endoscope based on said executable instructions;

an assessment of a severity of each said attribute of interest in said reduced set of images containing at least one attribute of interest.

34. The system of claim 33, where said assessment comprises calculating one of:

a score,

a rank,

a structured assessment comprising of one or more categories,

a structured assessment on a Likert scale, and

a relationship with one or more other images, wherein said relationship comprises less severe or more severe.

35. The system of claim 33, wherein said processing unit further performs the following in response to receiving a plurality of endoscopic images from said endoscope based on said executable instructions:

a derivation of a score for said reduced set of images containing at least one attribute of interest based on said severity of each said region of interest.

36. The system of claim 35, wherein said score comprises at least one of:

a Lewis score,

a Crohn'"'"'s Disease Endoscopy Index of Severity,

a Simple Endoscopic Score for Crohn'"'"'s Disease,

a Crohn'"'"'s Disease Activity Index, and

a rubric based on image appearance attributes, wherein said appearance attributes comprises one of;

lesion exudates, inflammation, color, and texture.

37. The system of claim 24, wherein said processing unit further performs the following in response to receiving a plurality of endoscopic images from said endoscope based on said executable instructions:

an identification of each of said plurality of endoscopic images to determine whether any of said plurality of endoscopic images is unusable for further processing; and

a removal of said unusable image from further processing.

38. The system according to claim 37, wherein said unusable image comprises at least one image of:

air bubbles,

food,

fecal matter,

normal tissue,

non-lesion, and

structures.

39. The system of claim 24, wherein said determination of whether at least one attribute of interest is present and said classification of said plurality of endoscopic images comprises using at least one of:

supervised machine learning and unsupervised machine learning.

40. The system of claim 24, wherein said determination of whether at least one attribute of interest is present comprises using at least one of:

statistical measures,

machine learning algorithms,

traditional classification techniques,

regression techniques,

feature vectors,

localized descriptors,

MPEG-7 visual descriptors,

edge features,

color histograms,

image statistics,

gradient statistics,

Haralick texture features,

dominant color descriptors,

edge histogram descriptors,

homogeneous texture descriptors,

spatial kernel weighting,

uniform grid sampling,

grid sampling with multiple scales,

local mode-seeking using mean shift,

generic lesion templates,

linear discriminate analysis,

logistic regression,

K-nearest neighbors,

relevance vector machines,

expectation maximation,

discrete wavelets, and

Gabor filters.

41. The system of claim 24, wherein said predetermined criterion comprises a measurement of at least one of:

color,

texture,

hue,

saturation,

intensity,

energy,

entropy,

maximum probability,

contrast,

inverse difference moment, and

correlation.

42. The system according to claim 24, wherein said classification of said plurality of endoscopic images comprises using at least one of:

meta methods,

boosting methods,

bagging methods,

voting,

weighted voting,

adaboost,

temporal consistency,

performing a second classification procedure on data neighboring said localized region of interest, and

Bayesian analysis.

43. The system of claim 24, wherein said endoscope comprises one of:

a wireless capsule endoscopy device,

a flexible endoscope,

a contact hysteroscope,

a flexible borescope,

a video borescope,

a rigid borescope,

a pipe borescope,

a GRIN lens endoscope, and

a fibroscope.

44. The method according to claim 24, wherein,

said plurality of endoscopic images are images taken within a gastrointestinal track;

and

said attribute of interest comprises an anatomic abnormality in said gastrointestinal track.

45. The method according to claim 44, wherein said anatomic abnormality comprises at least one of:

a lesion,

mucosal inflammation,

an erosion,

an ulcer,

submucosal inflammation,

a stricture,

a fistulae,

a perforation,

an erythema,

edema,

blood, and

a boundary organ.

46. The method according to claim 24, wherein said receiving a plurality of images from said endoscope comprises receiving images from one of:

a database of endoscopic images, and

in real-time from said endoscope.

47. A computer readable medium storing executable instructions for execution by a computer having memory, the medium storing instructions for:

receiving a plurality of endoscopic images;

processing each of said plurality of endoscopic images to determine whether at least one attribute of interest is present in each image that satisfies a predetermined criterion; and

classifying said plurality of endoscopic images into a reduced set of images each of which contains said at least one attribute of interest and a remainder set of images each of which is free from said at least one attribute of interest.