OPTIMAL, USER-FRIENDLY, OBJECT BACKGROUND SEPARATION

• US 20130301889A1
• Filed: 12/06/2011
• Published: 11/14/2013
• Est. Priority Date: 12/07/2010
• Status: Active Grant

• Current Assignee: The United States of America Represented by The Department of Veterans Affairs
• Original Assignee: University of Iowa Research Foundation
• Inventors: Abramoff, Michael, Quellec, Gwenole
• Granted Patent: US 10,140,699 B2
• Time in Patent Office: Days
• US Class Current: 382/128
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/30041 Eye; Retina; Ophthalmic
• CPC Class Codes: G06T 2207/30096 Tumor; Lesion
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/75 involving models
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images
• CPC Class Codes: G06V 40/193 Preprocessing; Feature extraction

Abstract

A method of identifying an object of interest can comprise obtaining first samples of an intensity distribution of one or more object of interest, obtaining second samples of an intensity distribution of confounder objects, transforming the first and second samples into an appropriate first space, performing dimension reduction on the transformed first and second samples, whereby the dimension reduction of the transformed first and second samples generates an object detector, transforming one or more of the digital images into the first space, performing dimension reduction on the transformed digital images, whereby the dimension reduction of the transformed digital images generates one or more reduced images, classifying one or more pixels of the one or more reduced images based on a comparison with the object detector, and identifying one ore more objects of interest from the classified pixels.

Claims

1. A method of identifying an object of interest in digital images, the method comprising:

a. obtaining first samples of an intensity distribution of one or more objects of interest in one or more of the digital images based upon one or more wavelength bands;

b. obtaining second samples of an intensity distribution of confounder objects in one or more of the digital images, at a predetermined frequency;

c. transforming the first and second samples into an appropriate first space;

d. performing dimension reduction on the transformed first and second samples, whereby the dimension reduction of the transformed first and second samples generates an object detector;

e. transforming one or more of the digital images into the first space;

f. performing dimension reduction on the transformed digital images, whereby the dimension reduction of the transformed digital images generates one or more reduced images;

g. classifying one or more pixels of the one or more reduced images based on a comparison with the object detector; and

h. identifying one or more objects of interest in the reduced digital images from the classified pixels.

2. The method of claim 1, further comprising combining the plurality of digital images into a single image.

3. The method of claim 1, further comprising identifying a disease based on the distribution of the object of interests in one or more of the plurality of digital images.

4. The method of claim 3, wherein the identification criteria comprises one or more of the presence of an object of interest, the number of objects of interest and the spatial distribution of the object of interest.

5. The method of claim 1, wherein one or more of the confounder objects comprise a negative object confounder.

6. The method of claim 1, wherein one or more of the confounder objects comprise a positive object confounder.

7. The method of claim 1, further comprising updating one or more of the first samples and the second samples by adding false positives of a classifier to a set of the samples.

8. The method of claim 1, wherein one ore more of the first samples of the intensity distribution of objects are obtained by mathematically modeling the object, and transforming the model along one or more axes, and then obtaining multiple intensity distributions from each perturbation.

9. The method of claim 1, wherein one ore more of the first samples of the intensity distribution of objects are obtained by sampling the intensity distribution of images in a region where a object has been detected by an external observer.

10. The method of claim 1, where the predetermined frequency is set based upon an area under a ROC curve.

11. The method of claim 1, wherein the transformation under step c) is a wavelet transformation.

12. The method of claim 1, wherein the transformation under step c) is an AM-FM transformation.

13. The method of claim 1, wherein the transformation under step c) is an identity transform.

14. The method of claim 1, wherein the dimension reduction under step d) is performed by estimating one or more principal components of a distribution obtained by performing a principal component analysis.

15. The method of claim 1, wherein the dimension reduction under step d) is performed by estimating one or more independent components of a distribution obtained by performing an independent component analysis.

16. The method of claim 1, wherein the dimension reduction under step d) is performed by estimating one or more factors of a distribution obtained by performing a factor analysis.

17. The method of claim 1, wherein the dimension reduction under step d) is performed by estimating one or more intrinsic mode functions of a distribution obtained by performing an empirical mode decomposition.

18. The method of claim 1, wherein the pixels under step g) are classified as of by:

locating one or more K samples that minimize a distance to pixels in a pre-defined neighborhood; and

classifying one or more pixels as one of abnormal and normal, using the distance to the K samples and a label associated with the K samples.

19. The method of claim 18, wherein a k-nearest neighbor algorithm based on kd-trees is used to locate the one or more K samples.

20. The method of claim 18, wherein a k-nearest neighbor algorithm based on box-decomposition trees is used is used to locate the one or more K samples.

21. The method of claim 18, wherein a lesion likelihood index L_p for a pixel p of neighborhood N_p is obtained from the K nearest samples (S_k)_{k=1 . . . K}, with labels (l_k)_{k=1 . . . k} equal to 1 for lesions and −

1 for negative lesion confounders, by the following formula L_pΣ

_k=1^Kl_k exp(−

∥

N_p−

S_k∥

) used to automatically classify the pixel p as abnormal or normal.

22. The method of claim 1, wherein the pixels are classified as normal or abnormal using one or more of a discriminant hyper plane and a plurality of discriminant hyper planes, in the space of reduced dimension obtained in step d).

23. The method of claim 1, wherein the objects of interest under step h) are obtained by clustering the connected pixels of interest, by means of morphological labeling.

24. The method of claim 1, wherein the intensity of one or more of the first samples, the second samples, and the pixels is normalized.

25. The method of claim 1, wherein the detected objects of interest are categorized into different types of objects of interest.

26. The method of claim 1, where the plurality of digital images comprises two-dimensional images.

27. The method of claim 1, where the plurality of digital images comprises three-dimensional (volumetric) images.

28. A method of identifying an object of interest in a plurality of digital images configured as an image group, where each of the digital images comprises a plurality of wavelength bands, the method comprising:

a) obtaining first samples of an intensity distribution of one or more objects of interest in the image group based upon one or more wavelength bands;

b) obtaining second samples of the intensity distribution of confounder objects in the image group, at a frequency high enough to affect a pre-defined performance metric;

c) transforming the first and second samples into an appropriate first space;

d) performing dimension reduction on the transformed first and second samples, whereby the dimension reduction of the transformed first and second samples generates an object detector;

e) transforming the image group into the first space;

f) projecting the transformed image group into the reduced dimension space obtained in step d) to generate a reduced image group;

g) classifying each pixel in the reduced image group with an appropriate neighborhood based on a comparison with the object detector; and

h) automatically identifying one or more objects of interest in the reduced image group from abnormal pixels based upon the comparison with the object detector.

29. A system comprising:

a memory for storing digital images; and

a processor in communication with the memory, the processor configured to;

a) obtain first samples of an intensity distribution of one or more objects of interest in one or more of the digital images based upon one or more wavelength bands;

b) obtain second samples of an intensity distribution of confounder objects in one or more of the digital images, at a predetermined frequency;

c) transform the first and second samples into an appropriate first space;

d) perform dimension reduction on the transformed first and second samples, whereby the dimension reduction of the transformed first and second samples generates an object detector;

e) transform one or more of the digital images into the first space;

f) perform dimension reduction on the transformed digital images, whereby the dimension reduction of the transformed digital images generates one or more reduced images;

g) classify one or more pixels of the one or more reduced images based on a comparison with the object detector; and

h) identify one or more objects of interest in the reduced digital images from the classified pixels.