Classification of polyps using learned image analysis

Patent metadata

Field	Value
Publication number	US-11666286-B2
Application number	US-202117305105-A
Country	US
Kind code	B2
Filing date	Jun 30, 2021
Priority date	Oct 7, 2016
Publication date	Jun 6, 2023
Grant date	Jun 6, 2023

Abstract

Computational techniques are applied to video images of polyps to extract features and patterns from different perspectives of a polyp. The extracted features and patterns are synthesized using registration techniques to remove artifacts and noise, thereby generating improved images for the polyp. The generated images of each polyp can be used for training and testing purposes, where a machine learning system separates two types of polyps.

First claim

The invention claimed is:

1. A system for classifying polyps, the system comprising:
a polyp imaging engine, the polyp imaging engine configured to:
generate, from images of at least one polyp taken from different perspectives, a new polyp image, the new polyp image having fewer reflection artifacts and occlusions than the images of the at least one polyp taken from different perspectives, and
compute, based on the generated new polyp image, a polyp surface model; and

a learning engine, the learning engine configured to:
apply linear subspace learning techniques and nonlinear subspace learning techniques to identify discriminate features in the polyp surface model, and
classify the at least one polyp based on the identified discriminate features.

2. The system of claim 1 , wherein classifying the at least one polyp based on the identified discriminate features includes classifying the at least one polyp as adenomatous or hyperplastic.

3. The system of claim 1 , further comprising a polyp image database comprising images of a plurality of polyps taken from different perspectives, wherein the at least one polyp is included in the plurality of polyps.

4. The system of claim 3 , wherein at least one of the images of the plurality of polyps taken from different perspectives includes images generated using Narrow Band Imaging (NBI).

5. The system of claim 4 wherein generating the images of the plurality of polyps taken from different perspectives using NBI comprises:
generating a first image of at least one of the plurality of polyps using a light source emitting at or about 415 nanometers, and
generating a second image of the at least one of the plurality of polyps using a light source emitting at or about 540 nanometers.

6. The system of claim 3 , wherein at least one of the images of the plurality of polyps taken from different perspectives comprises images generated using a white light (WL) source.

7. The system of claim 1 , wherein applying subspace learning techniques to identify discriminate features in the polyp surface model comprises:
developing sparse representations of the polyp to build polyp representations that are not affected by imagine artifacts and occlusions.

8. The system of claim 7 , wherein the developed sparse representations are utilized to:
estimate accurate decision functions and separation between statistical prototype of adenomas and non-adenomas polyps.

9. The system of claim 1 wherein the polyp imaging engine is further configured to:
generate color and texture data for images of the at least one polyp taken from different perspectives.

10. A method for classifying polyps, the method comprising:
generating, from images of at least one polyp taken from different perspectives, a new polyp image, the new polyp image having fewer reflection artifacts and occlusions than the images of the at least one polyp taken from different perspectives;
computing, based on the generated new polyp image, a polyp surface model; and
applying, by a learning engine, a linear subspace learning techniques and nonlinear subspace learning techniques to identify discriminate features in the polyp surface model, and
classifying the at least one polyp based on the identified discriminate features.

11. The method of claim 10 , wherein classifying the at least one polyp based on the identified discriminate features includes classifying the at least one polyp as adenomatous or hyperplastic.

12. The method of claim 10 , further comprising:
storing, in an image database, images of a plurality of polyps taken from different perspectives, wherein the at least one polyp is included in the plurality of polyps.

13. The method of claim 12 , further comprising:
generating at least one of the images of the plurality of polyps taken from different perspective using Narrow Band Imaging (NBI).

14. The method of claim 13 , wherein generating the images of the plurality of polyps taken from different perspectives using NBI comprises:
generating a first image of at least one of the plurality of polyps using a light source emitting at or about 415 nanometers, and
generating a second image of the at least one of the plurality of polyps using a light source emitting at or about 540 nanometers.

15. The method of claim 12 , wherein at least one of the images of the plurality of polyps taken from different perspectives comprises images generated using a white light (WL) source.

16. The method of claim 10 , further comprising:
developing sparse representations of the at least one polyp to build polyp representations that are not affected by imagine artifacts and occlusions.

17. The method of claim 16 , further comprising:
estimating accurate decision functions and separation between statistical prototype of adenomas and non-adenomas polyps.

18. The method of claim 10 , further comprising:
generating color and texture data for the images of the at least one polyp taken from different perspectives.

Assignees

• Baylor Res Institute
• Univ Wright State

Inventors

• Dassopoulos Themistocles
• Bourbakis Nikolaos

Classifications

• G06T11/10

  Texturing; Colouring; Generation of textures or colours (retouching, inpainting or scratch removal G06T5/77) · CPC title

• G06V2201/032

  of protuberances, polyps nodules, etc. · CPC title

• G06V10/16

  using multiple overlapping images; Image stitching · CPC title

• G06T2207/30004

  Biomedical image processing · CPC title

• G06F16/583

  using metadata automatically derived from the content · CPC title