Method and system for computer-aided lesion detection using information from multiple images

• US 6,075,879 A
• Filed: 10/26/1998
• Issued: 06/13/2000
• Est. Priority Date: 09/29/1993
• Status: Expired due to Term

• Current Assignee: Hologic Incorporated
• Original Assignee: R2 Technology, Inc. (Hologic Incorporated)
• Inventors: Roehrig, Jimmy R., Wang, Shih-Ping, Schneider, Alexander C.
• Primary Examiner(s): MEHTA, BHAVESH M
• Application Number:
  US09/178,901
• Time in Patent Office: 596 Days
• Field of Search: 382/128, 382/131, 382/132, 382/284, 382/173, 382/224, 378/37, 128/920, 128/922
• US Class Current: 382/132
• CPC Class Codes: G06T 2207/30068 Mammography; Breast
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G16H 30/20 for handling medical images, e.g. DICOM, HL7 or PACS
• CPC Class Codes: G16H 40/63 for local operation
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: H05G 1/60 Circuit arrangements for obtaining a series of X-ray photographs or for X-ray cinematography
• CPC Class Codes: Y10S 128/922 including image analysis

Abstract

A system, method, and computer program product for computer-aided detection of suspicious lesions in digital mammograms is described, wherein single-view feature vectors from a first digital mammogram are processed in a classification algorithm along with information computed from a plurality of related digital mammograms to assign an overall probability of suspiciousness to potentially suspicious lesions in the first digital mammogram. In one preferred embodiment, a greater probability of suspiciousness is determined where there are similar corresponding lesions in the first digital mammogram and in an alternate digital mammogram view of the same breast. In another preferred embodiment, a lesser probability of suspiciousness is found where there are symmetric lesions or structures located in the first digital mammogram and a digital mammogram of the opposite breast. In another preferred embodiment, a lesser probability of suspiciousness is found where there are similar lesions or structures located in the first digital mammogram and a digital mammogram of the same breast taken months or years earlier in time. In another preferred embodiment, the nipple location, which serves as a reference location point across different digital mammogram, is located using an algorithm that takes into account the chest wall, the skin line of the breast, and the general orientation of the fibrous breast tissue in the digital mammogram relative to the chest wall.

Claims

1. A method for detecting suspicious lesions in a breast using information from a first digital mammogram view of the breast and a second digital mammogram view of the breast, comprising the steps of:

locating a first set of potentially suspicious lesions in said first digital mammogram view, each of said first set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

locating a second set of potentially suspicious lesions in said second digital mammogram view;

computing a similarity metric between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions; and

classifying each of said first set of potentially suspicious lesions using information from the corresponding single-view feature vector and from the corresponding similarity metrics.

2. The method of claim 1, wherein said step of classifying each of said first set of potentially suspicious lesions is performed using a classification algorithm in which the single-view feature vector associated with that potentially suspicious lesion is processed along with the corresponding similarity metrics, and wherein a greater probability of suspiciousness is determined if that potentially suspicious lesion is highly similar to a potentially suspicious lesion in the second digital mammogram view.

3. The method of claim 2, each of said second set of potentially suspicious lesions also having a single-view feature vector, said single-view feature vectors associated with said first and second sets of potentially suspicious lesions comprising location information, wherein said step of computing a similarity metric comprises the step of comparing location information between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions.

4. The method of claim 3, wherein said location information includes a nipple distance corresponding to a distance of the potentially suspicious lesion from the nipple of the breast in its respective digital mammogram view.

5. The method of claim 4, further comprising the step of locating the nipple of the breast in each digital mammogram view, said step of locating the nipple comprising the steps of:

segmenting that digital mammogram view for determining the position of the chest wall of the breast;

determining a most prominent direction of lines in the breast tissue relative to the chest wall for that digital mammogram view;

segmenting that digital mammogram view for determining the location of the skin line of the breast; and

selecting the location of the nipple on that digital mammogram view as the furthest point along said skin line from said chest wall as measured along said most prominent direction of lines.

6. The method of claim 5, wherein said nipple distance is equal to the radial distance of the potentially suspicious lesion from the nipple in that digital mammogram view.

7. The method of claim 5, wherein said nipple distance is equal to the difference between (a) the distance of the nipple from the chest wall and (b) the distance of the potentially suspicious lesion from the chest wall.

8. The method of claim 5, wherein said single-view feature vector for each potentially suspicious lesion further comprises (1) an area metric related to the area of the potentially suspicious lesion, and (2) a contrast metric related to the contrast of the potentially suspicious lesion relative to the surrounding tissue area, and wherein said similarity metric is computed by computing a weighted distance in feature vector space between the potentially suspicious lesion in the first digital mammogram view and the potentially suspicious lesion in the second digital mammogram view.

9. The method of claim 8, wherein that potentially suspicious lesion is highly similar to the potentially suspicious lesion in the second digital mammogram view if said weighted distance in feature vector space has a low numerical value, wherein said weighted distance in feature vector space is a weighted Euclidean distance computed using the steps of:

determining a difference in nipple distance between the potentially suspicious lesion in the first digital mammogram view and the potentially suspicious lesion in the second digital mammogram view;

squaring said difference in nipple distance and weighting the result by a nipple distance weighting factor;

determining a difference in area metric between the potentially suspicious lesion in the first digital mammogram view and the potentially suspicious lesion in the second digital mammogram view;

squaring said difference in area metric and weighting the result by an area metric weighting factor; and

computing said weighted Euclidean distance by computing the square root of the sum of said weighted squared differences.

10. The method of claim 9, wherein said nipple distance weighting factor is selected to be at least approximately five times greater than said area metric weighting factor.

11. The method of claim 10, said single-view feature vectors each further comprising a spiculation metric corresponding to the spiculatedness of the potentially suspicious lesion, wherein said weighted distance in feature vector space is further computed using the steps of:

determining a difference in spiculation metric between the potentially suspicious lesion in the first digital mammogram view and the potentially suspicious lesion in the second digital mammogram view;

squaring said difference in spiculation metric and weighting the result by a spiculation metric weighting factor;

determining a difference in contrast metric between the potentially suspicious lesion in the first digital mammogram view and the potentially suspicious lesion in the second digital mammogram view;

squaring said difference in contrast metric and weighting the result by a contrast metric weighting factor; and

including said weighted squared spiculation and contrast metrics in said sum of said weighted squared differences.

12. A method for detecting suspicious lesions in a first breast using information from a first digital mammogram thereof and a second digital mammogram of an opposite breast, comprising the steps of:

locating a first set of potentially suspicious lesions in said first digital mammogram, each of said first set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

locating a second set of potentially suspicious lesions in said second digital mammogram;

computing a symmetry metric between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions; and

classifying each of said first set of potentially suspicious lesions using information from the corresponding single-view feature vector and from the corresponding symmetry metrics relating that potentially suspicious lesion to potentially suspicious lesions in the opposite breast.

13. The method of claim 12, wherein said step of classifying each of said first set of potentially suspicious lesions is performed using a classification algorithm in which the single-view feature vector associated with that potentially suspicious lesion is processed along with the corresponding symmetry metrics relating that potentially suspicious lesion to potentially suspicious lesions in the opposite breast, and wherein a lesser probability of suspiciousness is determined if that potentially suspicious lesion is highly symmetric with a potentially suspicious lesion in the opposite breast.

14. The method of claim 13, each of said second set of potentially suspicious lesions also having a single-view feature vector, said single-view feature vectors associated with said first and second sets of potentially suspicious lesions comprising location information, wherein said step of computing a similarity metric comprises the step of comparing location information between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions.

15. The method of claim 14, wherein said location information includes a nipple distance corresponding to a distance of the potentially suspicious lesion from the nipple of the breast in its respective digital mammogram.

16. The method of claim 15, further comprising the step of locating the nipple of the breast in each of said first and second digital mammograms, said step of locating the nipple in a respective digital mammogram comprising the steps of:

segmenting the digital mammogram for determining the position of the chest wall of the breast;

determining a most prominent direction of lines in the breast tissue relative to the chest wall for the digital mammogram;

segmenting the digital mammogram for determining the location of the skin line of the breast; and

selecting the location of the nipple on the digital mammogram as the furthest point along said skin line from said chest wall as measured along said most prominent direction of lines.

17. The method of claim 16, wherein said nipple distance is equal to the radial distance of the potentially suspicious lesion from the nipple in the digital mammogram.

18. The method of claim 16, wherein said nipple distance is equal to the difference between (a) the distance of the nipple from the chest wall and (b) the distance of the potentially suspicious lesion from the chest wall.

19. The method of claim 15, wherein said single-view feature vector for each potentially suspicious lesion further comprises an area metric related to the area of the potentially suspicious lesion, and a contrast metric related to the contrast of the potentially suspicious lesion relative to the surrounding tissue area, and wherein said symmetry metric is computed by computing a weighted distance in feature vector space between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the second digital mammogram.

20. The method of claim 19, wherein that potentially suspicious lesion is highly symmetric with the potentially suspicious lesion in the second digital mammogram if said weighted distance in feature vector space has a low numerical value, wherein said weighted distance in feature vector space is a weighted Euclidean distance computed using the steps of:

determining a difference in nipple distance between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the second digital mammogram;

squaring said difference in nipple distance and weighting the result by a nipple distance weighting factor;

determining a difference in area metric between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the second digital mammogram;

squaring said difference in area metric and weighting the result by an area metric weighting factor; and

computing said weighted Euclidean distance by computing the square root of the sum of said weighted squared differences.

21. The method of claim 20, wherein said nipple distance weighting factor is selected to be at least approximately five times greater than said area metric weighting factor.

22. The method of claim 21, said single-view feature vectors each further comprising a spiculation metric corresponding to the spiculatedness of the potentially suspicious lesion, wherein said weighted distance in feature vector space is further computed using the steps of:

determining a difference in spiculation metric between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the second digital mammogram;

squaring said difference in spiculation metric and weighting the result by a spiculation metric weighting factor;

determining a difference in contrast metric between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the second digital mammogram;

squaring said difference in contrast metric and weighting the result by a contrast metric weighting factor; and

including said weighted squared spiculation and contrast metrics in said sum of said weighted squared differences.

23. A method for detecting suspicious lesions in a first breast using information from a first digital mammogram thereof taken from a first view, a second digital mammogram thereof taken from a second view different than the first view, and a third digital mammogram taken of an opposite breast, comprising the steps of:

locating a first set of potentially suspicious lesions in said first digital mammogram, each of said first set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

locating a second set of potentially suspicious lesions in said second digital mammogram of the breast, each of said second set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

locating a third set of potentially suspicious lesions in said third digital mammogram, each of said third set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

computing a similarity metric between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions;

computing a symmetry metric between each of said first set of potentially suspicious lesions and each of said third set of potentially suspicious lesions; and

classifying each of said first set of potentially suspicious lesions using information from the (a) the corresponding single-view feature vector, (b) the corresponding similarity metrics relating that potentially suspicious lesion to potentially suspicious lesions in the different view of the same breast, and (c) the corresponding symmetry metrics relating that potentially suspicious lesion to potentially suspicious lesions in the opposite breast.

24. The method of claim 23, wherein said step of classifying each of said first set of potentially suspicious lesions is performed using a classification algorithm in which the single-view feature vector associated with that potentially suspicious lesion is processed along with the corresponding similarity and symmetry metrics, wherein a greater probability of suspiciousness is determined if that potentially suspicious lesion is highly similar to a potentially suspicious lesion in the second digital mammogram, and wherein a lesser probability of suspiciousness is determined if that potentially suspicious lesion is highly symmetric with a potentially suspicious lesion in the third digital mammogram.

25. The method of claim 24, said single-view feature vectors comprising location information, wherein said step of computing a similarity metric comprises the step of comparing location information between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions, and wherein said step of computing a symmetry metric comprises the step of comparing location information between each of said first set of potentially suspicious lesions and each of said third set of potentially suspicious lesions.

26. The method of claim 25, wherein said location information includes a nipple distance corresponding to a distance of the potentially suspicious lesion from the nipple of the breast in its respective digital mammogram.

27. The method of claim 26, wherein said single-view feature vector for each potentially suspicious lesion further comprises an area metric related to the area of the potentially suspicious lesion, and a contrast metric related to the contrast of the potentially suspicious lesion relative to the surrounding tissue area, wherein said similarity metric is computed by computing a weighted distance in feature vector space between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the second digital mammogram, and wherein symmetry metric is computed by computing a weighted distance in feature vector space between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the third digital mammogram.

28. The method of claim 27, wherein that potentially suspicious lesion is highly similar to the potentially suspicious lesion in the second digital mammogram if said weighted distance in feature vector space has a low numerical value, wherein said weighted distance in feature vector space is a weighted Euclidean distance computed using the steps of:

determining a difference in nipple distance between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the second digital mammogram;

squaring said difference in nipple distance and weighting the result by a similarity nipple distance weighting factor;

determining a difference in area metric between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the second digital mammogram;

squaring said difference in area metric and weighting the result by an similarity area metric weighting factor; and

computing said weighted Euclidean distance by computing the square root of the sum of said weighted squared differences.

29. The method of claim 28, wherein that potentially suspicious lesion is highly symmetric with the potentially suspicious lesion in the third digital mammogram if said weighted distance in feature vector space has a low numerical value, wherein said weighted distance in feature vector space is a weighted Euclidean distance computed using the steps of:

determining a difference in nipple distance between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the third digital mammogram;

squaring said difference in nipple distance and weighting the result by a symmetry nipple distance weighting factor;

determining a difference in area metric between the potentially suspicious lesion in the first digital mammogram and the potentially suspicious lesion in the third digital mammogram;

squaring said difference in area metric and weighting the result by an symmetry area metric weighting factor; and

computing said weighted Euclidean distance by computing the square root of the sum of said weighted squared differences.

30. The method of claim 29, wherein said similarity nipple distance weighting factor is approximately equal to said symmetry nipple distance weighting factor, and wherein said similarity area metric weighting factor is approximately equal to aid symmetry area metric weighting factor.

31. The method of claim 30, wherein said similarity nipple distance weighting factor is at least approximately five times greater than said similarity area metric weighting factor.

32. A computer program product for directing a computer-aided lesion detection apparatus to detect suspicious lesions in a breast using information from a first digital mammogram view of the breast and a second digital mammogram view of the breast, comprising:

computer code for locating a first set of potentially suspicious lesions in said first digital mammogram view, each of said first set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

computer code for locating a second set of potentially suspicious lesions in said second digital mammogram view;

computer code for computing a similarity metric between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions; and

computer code for classifying each of said first set of potentially suspicious lesions using information from the corresponding single-view feature vector and from the corresponding similarity metrics.

33. The computer program product of claim 32, wherein said computer code for classifying each of said first set of potentially suspicious lesions comprises computer code for performing a classification algorithm in which the single-view feature vector associated with that potentially suspicious lesion is processed along with the corresponding similarity metrics, and wherein a greater probability of suspiciousness is determined if that potentially suspicious lesion is highly similar to a potentially suspicious lesion in the second digital mammogram view.

34. A computer program product for directing a computer-aided lesion detection apparatus to detect suspicious lesions in a first breast using information from a first digital mammogram thereof and a second digital mammogram of an opposite breast, comprising:

computer code for locating a first set of potentially suspicious lesions in said first digital mammogram, each of said first set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

computer code for locating a second set of potentially suspicious lesions in said second digital mammogram;

computer code for computing a symmetry metric between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions; and

computer code for classifying each of said first set of potentially suspicious lesions using information from the corresponding single-view feature vector and from the corresponding symmetry metrics relating that potentially suspicious lesion to potentially suspicious lesions in the opposite breast.

35. The computer program product claim 34, wherein said computer code for classifying each of said first set of potentially suspicious lesions comprises computer code for performing a classification algorithm in which the single-view feature vector is processed along with the corresponding symmetry metrics relating that potentially suspicious lesion to potentially suspicious lesions in the opposite breast, and wherein a lesser probability of suspiciousness is determined if that potentially suspicious lesion is highly symmetric with a potentially suspicious lesion in the opposite breast.

36. A computer program product directing a computer-aided lesion detection apparatus to detect suspicious lesions in a first breast using information from a first digital mammogram thereof taken from a first view, a second digital mammogram thereof taken from a second view different than the first view, and a third digital mammogram taken of an opposite breast, comprising:

computer code for locating a first set of potentially suspicious lesions in said first digital mammogram, each of said first set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

computer code for locating a second set of potentially suspicious lesions in said second digital mammogram of the breast, each of said second set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

computer code for locating a third set of potentially suspicious lesions in said third digital mammogram, each of said third set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

computer code for computing a similarity metric between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions;

computer code for computing a symmetry metric between each of said first set of potentially suspicious lesions and each of said third set of potentially suspicious lesions; and

computer code for classifying each of said first set of potentially suspicious lesions using information from the (a) the corresponding single-view feature vector, (b) the corresponding similarity metrics relating that potentially suspicious lesion to potentially suspicious lesions in the different view of the same breast, and (c) the corresponding symmetry metrics relating that potentially suspicious lesion to potentially suspicious lesions in the opposite breast.

37. The computer program product of claim 36, wherein said computer code for classifying each of said first set of potentially suspicious lesions comprises computer code for performing a classification algorithm in which the single-view feature vector associated with that potentially suspicious lesion is processed along with the corresponding similarity and symmetry metrics, wherein a greater probability of suspiciousness is determined if that potentially suspicious lesion is highly similar to a potentially suspicious lesion in the second digital mammogram, and wherein a lesser probability of suspiciousness is determined if that potentially suspicious lesion is highly symmetric with a potentially suspicious lesion in the third digital mammogram.

38. A method for detecting suspicious lesions in a breast using information from a first digital mammogram thereof and a second digital mammogram thereof, comprising the steps of:

locating a first set of potentially suspicious lesions in said first digital mammogram, each of said first set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

locating a second set of potentially suspicious lesions in said second digital mammogram;

computing a similarity metric between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions; and

classifying each of said first set of potentially suspicious lesions using information from the corresponding single-view feature vector and from the corresponding similarity metrics;

wherein said first digital mammogram and said second digital mammogram correspond to mammograms of the breast taken at substantially different times, whereby change in the breast over time that may indicate one or more cancerous conditions may be detected.

39. The method of claim 38, wherein said step of classifying each of said first set of potentially suspicious lesions is performed using a classification algorithm in which the single-view feature vector associated with that potentially suspicious lesion is processed along with the corresponding similarity metrics, and wherein a lesser probability of suspiciousness is determined if that potentially suspicious lesion is highly similar to one of said second set of potentially suspicious lesions in said second digital mammogram.

40. The method of claim 39, wherein said first digital mammogram and said second digital mammogram are least 6 months apart in time.

41. The method of claim 40, wherein said first digital mammogram is taken prior to said second digital mammogram.

42. A computer program product for directing a computer-aided lesion detection apparatus to detect suspicious lesions in a breast using information from a first digital mammogram thereof and a second digital mammogram thereof, comprising:

computer code for locating a first set of potentially suspicious lesions in said first digital mammogram, each of said first set of potentially suspicious lesions having a single-view feature vector corresponding thereto;

computer code for locating a second set of potentially suspicious lesions in said second digital mammogram;

computer code for computing a similarity metric between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions; and

computer code for classifying each of said first set of potentially suspicious lesions using information from the corresponding single-view feature vector and from the corresponding similarity metrics;

wherein said first digital mammogram and said second digital mammogram correspond to mammograms of the breast taken at substantially different times, whereby change in the breast over time that may indicate one or more cancerous conditions may be detected.

43. The computer program product of claim 42, wherein said computer code for classifying each of said first set of potentially suspicious lesions comprises computer code for performing a classification algorithm in which the single-view feature vector associated with that potentially suspicious lesion is processed along with the corresponding similarity metrics, and wherein a lesser probability of suspiciousness is determined if that potentially suspicious lesion is highly similar to one of said second set of potentially suspicious lesions in said second digital mammogram.

44. The computer program product of claim 42, wherein said first digital mammogram and said second digital mammogram are least 6 months apart in time.

45. The computer program product of claim 44, wherein said first digital mammogram is taken prior to said second digital mammogram.

46. A method for detecting suspicious lesions in a breast using information from a first digital mammogram thereof taken from a first view, a second digital mammogram thereof taken from a second view different than the first view, and a third digital mammogram thereof taken at a time prior to the time at which the first digital mammogram was taken, comprising the steps of:

locating potentially suspicious lesions in each of said first, second, and third digital mammograms to produce a first, second, and third set of potentially suspicious lesions, respectively, each of said potentially suspicious lesions having a single-view feature vector corresponding thereto;

computing a similarity metric between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions;

computing a similarity metric between each of said first set of potentially suspicious lesions and each of said third set of potentially suspicious lesions; and

classifying each of said first set of potentially suspicious lesions using information from the (a) the corresponding single-view feature vector, (b) the corresponding similarity metrics relating that potentially suspicious lesion to potentially suspicious lesions in the second digital mammogram, and (c) the corresponding similarity metrics relating that potentially suspicious lesion to potentially suspicious lesions in the third digital mammogram;

wherein said step of classifying each of said first set of potentially suspicious lesions is performed using a classification algorithm in which the single-view feature vector associated with that potentially suspicious lesion is processed along with the corresponding similarity metrics, wherein a greater probability of suspiciousness is determined if that potentially suspicious lesion is highly similar to a potentially suspicious lesion in the second digital mammogram view, and wherein a lesser probability of suspiciousness is determined if that potentially suspicious lesion is highly similar to a potentially suspicious lesion in the third digital mammogram view.

47. A method for detecting suspicious lesions in a breast using information from a first digital mammogram thereof taken from a first view, a second digital mammogram thereof taken at a time prior to the time at which the first digital mammogram was taken, and a third digital mammogram of the opposing breast, comprising the steps of:

locating potentially suspicious lesions in each of said first, second, and third digital mammograms to produce a first, second, and third set of potentially suspicious lesions, respectively, each of said potentially suspicious lesions having a single-view feature vector corresponding thereto;

computing a similarity metric between each of said first set of potentially suspicious lesions and each of said second set of potentially suspicious lesions;

computing a symmetry metric between each of said first set of potentially suspicious lesions and each of said third set of potentially suspicious lesions; and

classifying each of said first set of potentially suspicious lesions using information from the (a) the corresponding single-view feature vector, (b) the corresponding similarity metrics relating that potentially suspicious lesion to potentially suspicious lesions in the second digital mammogram, and (c) the corresponding symmetry metrics relating that potentially suspicious lesion to potentially suspicious lesions in the third digital mammogram;

wherein said step of classifying each of said first set of potentially suspicious lesions is performed using a classification algorithm in which the single-view feature vector associated with that potentially suspicious lesion is processed along with the corresponding symmetry and similarity metrics, wherein a lesser probability of suspiciousness is determined if that potentially suspicious lesion is highly similar to a potentially suspicious lesion in the second digital mammogram view, and wherein a lesser probability of suspiciousness is determined if that potentially suspicious lesion is highly symmetric with a potentially suspicious lesion in the third digital mammogram view.