Method and system for automated detection of clustered microcalcifications from digital mammograms

• US 6,205,236 B1
• Filed: 10/12/1999
• Issued: 03/20/2001
• Est. Priority Date: 08/28/1997
• Status: Expired due to Fees

• Current Assignee: iCAD Incorporated
• Original Assignee: Qualia Computing Inc. (iCAD Incorporated)
• Inventors: Rosenstengel, John E., Rogers, Steven K., Amburn, Philip, DeSimio, Martin P., Broussard, Randy P., Ochoa, Edward M., Berkey, Telford S., Hoffmeister, Jeffrey W., Rathbun, Thomas F.
• Primary Examiner(s): Johns, Andrew W.
• Application Number:
  US09/416,437
• Time in Patent Office: 525 Days
• Field of Search: 382/128, 382/132, 382/155, 382/156, 382/206, 382/225, 382/256, 382/257, 382/260, 382/270, 382/272, 128/922, 706/13, 706/924, 378/37
• US Class Current: 382/132
• CPC Class Codes: A61B 6/502 for diagnosis of breast, i.e. mammography
• CPC Class Codes: B25J 15/04 with provision for the remote detachment or exchange of the head or parts thereof
• CPC Class Codes: B29C 2791/006 Using vacuum
• CPC Class Codes: B29C 2791/007 Using fluid under pressure
• CPC Class Codes: B29C 49/0021 using moulds or mould parts movable in a closed path, e.g. mounted on movable endless supports
• CPC Class Codes: B29L 2023/18 Pleated or corrugated hoses
• CPC Class Codes: G06F 18/41 Interactive pattern learning with a human teacher
• CPC Class Codes: G06T 5/20 using local operators
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06V 10/28 Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
• CPC Class Codes: G06V 10/443 by matching or filtering
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images
• CPC Class Codes: Y10S 128/922 including image analysis

Abstract

A method and system for detecting and displaying clustered microcalcifications in a digital mammogram, wherein a single digital mammogram is first automatically cropped to a breast area sub-image which is then processed by means of an optimized Difference of Gaussians filter to enhance the appearance of potential microcalcifications in the sub-image. The potential microcalcifications are thresholded clusters are detected, features are computed for the detected clusters, and the clusters are classified as either suspicious or not suspicious by means of a neural network. Thresholding is preferably by sloping local thresholding but may also be performed by global and dual-local thresholding. The locations in the original digital mammogram of the suspicious detected clustered microcalcifications are indicated. Parameters for use in the detection and thresholding portions of the system are computer-optimized by means of a genetic algorithm. The results of the system are optimally combined with a radiologist'"'"'s observation of the original mammogram by combining the observations with the results, after the radiologist has first accepted or rejected individual detections reported by the system.

Claims

1. A method for detecting microcalcifications, prior to classification of detections, in a digital mammogram image by digital image processing in screening mammography comprising the steps of:

using an optimizing algorithm and a database of training images to obtain an optimized parameter value; and

applying a filtering algorithm using said optimized parameter to said digital mammogram image to produce a filtered image in which the appearance of microcalcifications has been enhanced relative to background information.

2. The method according to claim 1 wherein said step of using an optimizing algorithm comprises:

producing an optimized parameter value t where t represents an average microcalcification target size.

3. The method according to claim 1 wherein said step of using an optimizing algorithm comprises:

using a genetic algorithm.

4. The method according to claim 3 wherein using said genetic algorithm comprises the step of:

iteratively searching a search space comprising said optimized parameter value to maximize a predetermined objective fitness function.

5. The method according to claim 4 wherein said step of iteratively searching further comprises the step of:

iteratively searching said search space until said fitness function converges.

6. The method according to claim 4 wherein said step of iteratively searching further comprises the step of:

iteratively searching said search space for a predetermined number of iterations.

7. The method according to claim 4 wherein using said genetic algorithm further comprises the step of:

using a simplex method to identify said optimized parameter value which minimizes a predetermined cost function.

8. A method for automated clustered microcalcification detection, prior to classification of detections, by digital image processing in screening mammography, comprising:

storing a digital representation of a mammogram;

using an optimizing algorithm and a database of training images to obtain optimized parameter values;

applying a filtering algorithm using said optimized parameters to said digital representation to obtain a filtered image essentially comprising suspected microcalcifications;

obtaining from said filtered image an image essentially comprising single-pixel representations of said suspected microcalcifications; and

grouping said single-pixel representations into clusters using said optimized parameters.

9. The method according to claim 8, wherein said step of using an optimizing algorithm comprises:

using a genetic algorithm.

10. The method according to claim 9, wherein:

said step of using a genetic algorithm comprises obtaining a d_nn value and a μ

Cs_min value where d_nn represents a nearest neighbor distance and μ

Cs_min represents a number of detected microcalcifications; and

said step of grouping groups single-pixel representations into clusters which represent microcalcifications that are within the distance d_nn of μ

Cs_min other microcalcifications.

11. The method according to claim 10, where in said genetic algorithm further comprises the step of:

iteratively searching a solution space containing possible values for d_nn and μ

Cs_min to identify sets of values in which at least one fitness function is maximized.

12. The method according to claim 11, wherein said genetic algorithm further comprises the step of:

using a simplex method to identify at least one of said sets of values in which a cost function is minimized.

13. A method for automated clustered microcalcification detection in a mammogram image by digital image processing in screening mammography, comprising:

locating clusters of microcalcifications in said mammogram image;

extracting features of said clusters of microcalcifications; and

classifying said clusters of microcalcifications as suspicious or non-suspicious using said features in a multi-layer perceptron neural network;

wherein said step of classifying, comprises the steps of;

multiplying at least one of said features by a weight w_ij, where i is an index representing the i^th feature vector element of a feature vector x having N elements and j is an index representing a j^th first layer node;

using first layer nodes of said multi-layer perceptron neural network to calculate first layer outputs ƒ

_j, said first layer outputs ƒ

_j, being calculated according to the function;

$f_j=\tanh <br><br>\left(\underset{i=1}{\overset{N}{\sum <br><br>}}<br><br>\left(w_{i,j}\times <br><br>x_i\right)\right)$

where x_i comprises a computed feature vector element;

using said first layer outputs in a smoothly varying output function to obtain a series of resulting values; and

identifying a cluster associated with one of said resulting values as suspicious if said one resulting value satisfies a predetermined relationship with respect to a threshold value, or non-suspicious otherwise.

14. The method according to claim 13, wherein said step of classifying further comprises the steps of:

multiplying at least one of said first layer outputs ƒ

_j by a second weight u_j,k; and

using a result of said multiplying step as an input to at least one output node, the output of said at least one output node being calculated according to the function;

$z_k<br><br>\left(y\right)=\tanh <br><br>\left(\underset{j=1}{\overset{J}{\sum <br><br>}}<br><br>\left(u_{j,k}\times <br><br>y_j\right)\right)$

 

where y_j=ƒ

(x), k is an index representing the k^th output node, and J is the number of first layer outputs to be multiplied.

15. A method for automated clustered microcalcification detection in a digital representation of a mammogram by digital image processing, comprising:

filtering said digital representation to obtain a filtered image comprising filtered image pixels comprising suspected microcalcifications;

determining, for each said filtered image pixel, a threshold value derived from a predetermined portion of said digital representation, said predetermined portion corresponding to said filtered image pixel; and

thresholding each said filtered image pixel with said threshold value derived from said predetermined portion of said digital representation corresponding to said filtered image pixel to obtain a thresholded image comprising microcalcifications.

16. The method according to claim 15 wherein said step of determining comprises:

centering a window around a pixel in said digital representation, said pixel in said digital representation corresponding to said filtered image pixel;

computing the mean μ and

standard deviation σ

of the pixels within said window; and

computing said threshold value according to the function;

17. The method according to claim 15 wherein:

said step of filtering comprises filtering said digital representation with a difference of Gaussians filter.

18. The method according to claim 17 wherein said step of determining comprises:

centering a window around a pixel in said digital representation, said pixel in said digital representation corresponding to said filtered image pixel;

computing the mean μ and

standard deviation σ

of the pixels within said window; and

computing said threshold value according to the function;

19. The method according to claim 15 wherein said predetermined portion of said digital representation corresponding to said filtered image pixel is a window of pixels centered around a pixel in said digital representation, said pixel in said digital representation corresponding to said filtered image pixel.

20. The method according to claim 19 wherein said window of pixels is rectangular.

21. The method according to claim 19 wherein said window of pixels is square.

22. The method according to claim 19 wherein said window of pixels is ellipsoidal.

23. The method according to claim 19 wherein said window of pixels is circular.

24. The method according to claim 19 wherein said window of pixels is polygonal.

25. The method according to claim 19 wherein said window of pixels is hexagonal.

26. A system for detecting microcalcifications, prior to classification of detections, in a digital mammogram image by digital image processing in screening mammography comprising:

means using an optimizing algorithm and a database of training images to obtain an optimized parameter value; and

means for applying a filtering algorithm using said optimized parameter to said digital mammogram image to produce a filtered image in which the appearance of microcalcifications has been enhanced relative to background information.

27. The system according to claim 26 wherein said means using an optimizing algorithm comprises:

means for producing an optimized parameter value t where t represents an average microcalcification target size.

28. The system according to claim 27 wherein said means using an optimizing algorithm comprises:

means using a genetic algorithm.

29. The system according to claim 28 wherein said means using said genetic algorithm comprises:

means for iteratively searching a search space comprising said optimized parameter value to maximize a predetermined objective fitness function.

30. The system according to claim 29 wherein said means for iteratively searching further comprises:

means for iteratively searching said search space until said fitness function converges.

31. The system according to claim 29 wherein said means for iteratively searching further comprises:

means for iteratively searching said search space for a predetermined number of iterations.

32. The system according to claim 29 wherein said means using said genetic algorithm further comprises:

means using a simplex method to identify said optimized parameter value which minimizes a predetermined cost function.

33. An apparatus for automated clustered microcalcification detection, prior to classification of detections, by digital image processing in screening mammography comprising:

means for storing a digital representation of a mammogram;

means for optimizing parameter values using an optimizing algorithm and a database of training images;

means for applying a filtering algorithm, using said optimized parameters, to said digital representation to obtain a filtered image essentially comprising suspected microcalcifications;

means for obtaining from said filtered image an image essentially comprising single-pixel representations of said suspected microcalcifications; and

means for grouping said single-pixel representations into clusters using said optimized parameters.

34. The apparatus according to claim 33 wherein said optimizing algorithm is a genetic algorithm.

35. The apparatus according to claim 34 wherein said means for optimizing parameter values comprises:

means for obtaining a d_nn value and a μ

Cs_min value where d_nn represents a nearest neighbor distance and μ

Cs_min represents a number of detected microcalcifications; and

wherein said means for grouping groups said single-pixel representations into clusters which represent microcalcifications that are within the distance d_nn of μ

Cs_min other microcalcifications.

36. The apparatus according to claim 35 wherein said means for optimizing parameter values further comprises:

means for iteratively searching a solution space containing possible values for d_nn and μ

Cs_min to identify sets of values in which at least one fitness function is maximized.

37. The apparatus according to claim 36 wherein said means for optimizing parameter values further comprises:

means for using a simplex method to identify at least one of said sets of values in which a cost function is minimized.

38. An apparatus for automated clustered microcalcification detection in a mammogram image by digital image processing in screening mammography comprising:

means for locating clusters of microcalcifications in said digital representation;

means for extracting features of said clusters of microcalcifications; and

a classifier for classifying said clusters of microcalcifications as suspicious or non-suspicious using said features in a multi-layer perceptron neural network, said classifier further comprising;

means for multiplying at least one of said features by a weight w_ij, where i is an index representing the ith feature vector element of a feature vector x having N elements and j is an index representing a jth first layer node;

first layer nodes for calculating first layer outputs ƒ

_j according to the function;

$f_j=\tanh <br><br>\left(\underset{i=1}{\overset{N}{\sum <br><br>}}<br><br>\left(w_{i,j}\times <br><br>x_i\right)\right)$

where x_i comprises a computed feature vector element; and

smoothly varying output means operatively connected to said first layer nodes to produce a resulting value; and

means for identifying a cluster associated with said resulting value as suspicious if said resulting value satisfies a predetermined relationship with respect to a threshold value, or non-suspicious otherwise.

39. The apparatus according to claim 38, wherein said classifier further comprises:

means for multiplying at least one of said first layer outputs ƒ

_j by a second weight u_j,k; and

at least one output node for calculating outputs z_k, where k is an index representing a k^th output node, according to the function;

$z_k<br><br>\left(y\right)=\tanh <br><br>\left(\underset{j=1}{\overset{J}{\sum <br><br>}}<br><br>\left(u_{j,k}\times <br><br>y_j\right)\right)$

where y_j=ƒ

_j(x) and J is the number of first layer outputs to be multiplied.

40. A system for automated clustered microcalcification detection in a digital representation of a mammogram by digital image processing comprising:

a filter for filtering said digital representation to generate a filtered image comprising filtered image pixels comprising suspected microcalcifications;

means for determining, from said digital representation, a threshold value for each said filtered image pixel; and

thresholding means for thresholding each said filtered image pixel with its respective threshold value to generate a thresholded image comprising microcalcifications.

41. The system according to claim 40 wherein said means for determining comprises:

means for generating a value representative of a predetermined set of pixels distributed about a pixel in said digital representation corresponding to said filtered image pixel; and

means for computing said threshold value as a function of said representative value.

42. The system according to claim 41 wherein:

said means for generating comprises means for computing the mean μ and

the standard deviation σ

of said predetermined set of pixels.

43. The system according to claim 42 wherein:

said means for computing computes said threshold value according to the function;

44. The system according to claim 43 wherein:

said predetermined set of pixels comprises a rectangular window centered about said pixel in said digital representation.

45. The system according to claim 43 wherein:

said predetermined set of pixels comprises a square window centered about said pixel in said digital representation.

46. The system according to claim 43 wherein:

said predetermined set of pixels comprises an ellipsoidal window centered about said pixel in said digital representation.

47. The system according to claim 43 wherein:

said predetermined set of pixels comprises a circular window centered about said pixel in said digital representation.

48. The system according to claim 43 wherein:

said predetermined set of pixels comprises a polygonal window centered about said pixel in said digital representation.

49. The system according to claim 43 wherein:

said predetermined set of pixels comprises a hexagonal window centered about said pixel in said digital representation.

50. The system according to claim 43 wherein:

said filter comprises a difference of Gaussians filter.