Breast skin line detection in radiographic images

• US 8,675,934 B2
• Filed: 06/24/2011
• Issued: 03/18/2014
• Est. Priority Date: 04/30/2010
• Status: Active Grant

• Current Assignee: iCAD Incorporated
• Original Assignee: VuCOMP, Inc.
• Inventors: Wehnes, Jeffrey C., Pike, James H., Monaco, James P.
• Primary Examiner(s): TUCKER, WESLEY J
• Application Number:
  US13/168,614
• Publication Number:
  US 20110274327A1
• Time in Patent Office: 998 Days
• Field of Search: 382/128, 382130-133, 382/266
• US Class Current: 382/128
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06T 2207/10116 X-ray image
• CPC Class Codes: G06T 2207/20016 Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
• CPC Class Codes: G06T 2207/30068 Mammography; Breast
• CPC Class Codes: G06T 2207/30096 Tumor; Lesion
• CPC Class Codes: G06T 5/10 using non-spatial domain filtering
• CPC Class Codes: G06T 5/70 Denoising; Smoothing
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling
• CPC Class Codes: G06V 10/454 Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]

Abstract

An image segmentation embodiment comprises applying a second derivative operator to the pixels of an image, growing a set of contours using seeding grid points as potential contour starting points, determining a contour strength vector for each of the contour pixels, generating a partial ellipse representing an estimated location of an object in the image, dividing the partial ellipse into a plurality of support sectors with control points, determining a contour strength and position for each contour, adjusting a position of each sector control point based on the contour positions weighted by the contour strengths of the contours centered in the respective sector, fitting the partial ellipse to the adjusted positions of the control points, and generating a segmentation mask of the object based on the partial fitted ellipse.

Claims

1. A method for segmenting an image comprising pixels, the method comprising:

applying a second derivative operator to the pixels to derive first and second eigenvalues and an orientation angle for each of the pixels;

generating a seeding grid comprising an array of a subset of the pixels;

growing a set of contours using the subset of pixels as potential contour starting points;

pruning the set of contours to generate a pruned set of contours;

determining a contour strength vector for each of the contour pixels based on a maximum positive one of the eigenvalues and the orientation angle of the respective contour pixel;

generating a partial ellipse representing an estimated location of an object in the image by combining the contour strength vectors with a set of directional operators;

dividing the partial ellipse into a plurality of support sectors, each sector including a respective control point on the partial ellipse;

for each of the pruned contours centered in a respective sector, setting a contour strength to be a maximum of dot products of the respective contour strength vectors with a normal to the partial ellipse at the respective sector'"'"'s control point, and setting a contour position to be a location of the maximum product;

adjusting a position of each sector control point based on the contour positions weighted by the contour strengths of the pruned contours centered in the respective sector;

fitting the partial ellipse to the adjusted positions of the control points to generate a partial fitted ellipse; and

generating a segmentation mask of the object based on the partial fitted ellipse.

2. The method of claim 1, further comprising subsampling and smoothing the image before the applying the second derivative operator.

3. The method of claim 1, wherein a pixel spacing of the seeding grid is four pixels.

4. The method of claim 1, wherein the growing the set of contours further comprises, for each pixel in the subset of pixels, using the respective pixel as a contour starting point if its eigenvalues meet positive maximum eigenvalue criteria.

5. The method of claim 4, wherein the growing the set of contours further comprises, for each contour starting point, testing a neighboring pixel orthogonal to the starting point orientation angle, and including the neighboring pixel with the contour starting point if it meets the positive maximum eigenvalue criteria.

6. The method of claim 1, wherein the pruning comprises filtering the set of contours using at least one of an intensity profile filter and a directionality filter.

7. The method of claim 1, further comprising repeating the dividing the partial ellipse, the setting the contour strength and position for each of the pruned contours, the adjusting the position of each control point, and the fitting the partial ellipse steps using the partial fitted ellipse.

8. The method of claim 1, wherein the image is a mammogram image, wherein the object is a breast, and wherein the adjusted ellipse represents a breast skin line.

9. The method of claim 8, further comprising:

generating a straight-line model of an upper skin line of the breast, and

combining the straight-line model with the adjusted ellipse before the generating the segmentation mask.

10. The method of claim 9, further comprising generating a straight-line model of a pectoral line in the image, and using the pectoral line model in the segmentation mask generation.

11. A system for segmenting an image comprising pixels, the system comprising:

a curvature image generator applying a second derivative operator to the pixels to derive first and second eigenvalues and an orientation angle for each of the pixels;

a seeding grid generator generating a seeding grid comprising an array of a subset of the pixels;

a contour generator growing a set of contours using the subset of pixels as potential contour starting points;

a contour pruner pruning the set of contours to generate a pruned set of contours;

a vector generator determining a contour strength vector for each of the contour pixels based on a maximum positive one of the eigenvalues and the orientation angle of the respective contour pixel;

an ellipse generator generating a partial ellipse representing an estimated location of an object in the image by combining the contour strength vectors with a set of directional operators;

a sector divider dividing the partial ellipse into a plurality of support sectors, each sector including a respective control point on the partial ellipse;

a contour strength detector setting, for each of the pruned contours centered in a respective sector, a contour strength to be a maximum of dot products of the respective contour strength vectors with a normal to the partial ellipse at the respective sector'"'"'s control point, and setting, for each of the pruned contours centered in the respective sector, a contour position to be a location of the maximum product;

a control point adjuster adjusting a position of each sector control point based on the contour positions weighted by the contour strengths of the pruned contours centered in the respective sector;

an ellipse adjuster fitting the partial ellipse to the adjusted positions of the control points to generate a partial fitted ellipse; and

a segmentation mask generator generating a segmentation mask of the object based on the partial fitted ellipse.

12. A computer program product for segmenting an image, the computer program product having a non-transitory computer-readable medium with a computer program embodied thereon, the computer program comprising:

computer program code for applying a second derivative operator to the pixels to derive first and second eigenvalues and an orientation angle for each of the pixels;

computer program code for generating a seeding grid comprising an array of a subset of the pixels;

computer program code for growing a set of contours using the subset of pixels as potential contour starting points;

computer program code for pruning the set of contours to generate a pruned set of contours;

computer program code for determining a contour strength vector for each of the contour pixels based on a maximum positive one of the eigenvalues and the orientation angle of the respective contour pixel;

computer program code for generating a partial ellipse representing an estimated location of an object in the image by combining the contour strength vectors with a set of directional operators;

computer program code for dividing the partial ellipse into a plurality of support sectors, each sector including a respective control point on the partial ellipse;

computer program code for setting, for each of the pruned contours centered in a respective sector, a contour strength to be a maximum of dot products of the respective contour strength vectors with a normal to the partial ellipse at the respective sector'"'"'s control point, and for setting, for each of the pruned contours centered in a respective sector, a contour position to be a location of the maximum product;

computer program code for adjusting a position of each sector control point based on the contour positions weighted by the contour strengths of the pruned contours centered in the respective sector;

computer program code for fitting the partial ellipse to the adjusted positions of the control points to generate a partial fitted ellipse; and

computer program code for generating a segmentation mask of the object based on the partial fitted ellipse.

13. The computer program product of claim 12, further comprising computer program code for subsampling and smoothing the image before the applying the second derivative operator.

14. The computer program product of claim 12, wherein a pixel spacing of the seeding grid is four pixels.

15. The computer program product of claim 12, wherein the computer program code for growing the set of contours further comprises computer program code for using, for each pixel in the subset of pixels, the respective pixel as a contour starting point if its eigenvalues meet positive maximum eigenvalue criteria.

16. The computer program product of claim 15, wherein the computer program code for growing the set of contours further comprises, computer program code for testing, for each contour starting point, a neighboring pixel orthogonal to the starting point orientation angle, and computer program code for including the neighboring pixel with the contour starting point if it meets the positive maximum eigenvalue criteria.

17. The computer program product of claim 12, wherein the computer program code for pruning comprises computer program code for filtering the set of contours using at least one of an intensity profile filter and a directionality filter.

18. The computer program product of claim 12, further comprising computer program code for repeating the dividing the partial ellipse, the setting the contour strength and position for each of the pruned contours, the adjusting the position of each control point, and the fitting the partial ellipse steps using the partial fitted ellipse.

19. The computer program product of claim 12, wherein the image is a mammogram image, wherein the object is a breast, and wherein the adjusted ellipse represents a breast skin line.

20. The computer program product of claim 19, further comprising:

computer program code for generating a straight-line model of an upper skin line of the breast, and

computer program code for combining the straight-line model with the adjusted ellipse before the generating the segmentation mask.

21. The computer program product of claim 20, further comprising computer program code for generating a straight-line model of a pectoral line in the image, and for using the pectoral line model in the segmentation mask generation.