Granted Patent B2

US 8,260,014 · App. 12/102,144 · Granted Sep 4, 2012

Abstract

Claims (49)

1. A method of microcalcification detection in mammographic images, comprising:

forming a plurality of true microcalcification (mcc) clusters from a plurality of cancerous mammographic images that pass a mcc ground truth measure test;

forming a plurality of normal clusters from a plurality of normal mammographic images;

gathering spot and cluster features from said true mcc clusters and normal clusters;

extracting linear structure features in said cancerous mammographic images and normal mammographic images guided by said true mcc clusters and said normal clusters;

using said gathered spot and cluster features and said extracted linear structure features in mcc detector training; and

applying said mcc detector to mammographic test images which are under testing for cancerous lesions.

2. A method as in claim 1 , wherein said mcc ground truth measure test determines the degree of importance or involvement of a cancerous mammographic image or a true mcc cluster in mcc detection training.

3. A method as in claim 1 , wherein said extracting linear structure features comprises:

extracting rotateable bands in said cancerous mammographic images and normal mammographic images; and

processing said rotateable bands to generate linear structure features.

4. A method as in claim 1 , wherein said forming of true clusters comprises:

performing a cluster ground truth region test; and

randomly removing clusters within the region according to the test results.

5. A method as in claim 4 , wherein the test comprises comparing the number of clusters within the region to a threshold and randomly removing some of the clusters if the number of cluster is above the threshold.

6. A method as in claim 4 , wherein the test comprises comparing a size of a ground truth region to a size of the object of the image.

7. A method as in claim 3 , wherein the processing comprises determining an ensemble average curve ratio and a maximum relative magnitude of ensemble average curves to identify a linear structure feature.

8. A method as in claim 3 , wherein the processing comprises determining whether lines of the bands form a peak in Hough parameter space and determining the angle spread of the lines of the bands in Hough parameter space.

9. A method of microcalcification detection with cluster reduction in mammographic images, comprising:

forming a plurality of true microcalcification (mcc) clusters from a plurality of cancerous mammographic images that pass a mcc ground truth measure test;

forming a plurality of normal clusters from a plurality of normal mammographic images;

gathering spot and cluster features from said true mcc clusters and normal clusters;

extracting linear structure features in said cancerous mammographic images and normal mammographic images guided by said true mcc clusters and said normal clusters;

generating cluster reduction rules from distributions of said linear structure features;

using said spot and cluster features, said linear structure features and said cluster reduction rules in mcc detector training; and

applying said mcc detector and said cluster reduction rules to mammographic test images which are under testing for cancerous lesions.

10. A method as in claim 9 , wherein cluster reduction rules are cascade rules that confirm an association of a cluster with linear structures.

11. A method as in claim 9 , wherein using said cluster reduction rules in mcc detector training removes a certain number of clusters if they are confirmed as being associated with linear structures.

12. A method as in claim 9 , wherein applying said cluster reduction rules mammographic to test images removes a certain number of clusters if they are confirmed as being associated with linear structures.

13. A method as recited in claim 9 , wherein said generating comprises:

determining ensemble average curve statistics containing an ensemble average curve ratio and a maximum relative magnitude of ensemble average curves for cluster candidates;

tagging cluster candidates as linear, non-linear or uncertain responsive to the ensemble average curve statistics;

applying a Hough transform to an uncertain cluster candidate;

comparing a Hough spread angle to a spread angle threshold;

comparing a Hough normalized maximum magnitude to a normalized maximum magnitude threshold;

tagging uncertain clusters as linear or non-linear responsive to the comparisons; and

changing a cluster tagged as linear to non-linear when a ring hit threshold is exceeded.

14. A method of microcalcification detection incorporating linear structure features in mammographic images, comprising:

forming a plurality of true microcalcification (mcc) clusters from a plurality of cancerous mammographic images;

forming a plurality of normal clusters from a plurality of normal mammographic images;

gathering spot and cluster features from said true mcc clusters and normal clusters;

extracting linear structure features in said cancerous mammographic images and normal mammographic images guided by said true mcc clusters and said normal clusters;

using said gathered spot and cluster features and said extracted linear structure features in mcc detector training; and

applying said mcc detector to mammographic test images which are under testing for cancerous lesions.

15. An apparatus, comprising:

a display for displaying a result; and

a computer detecting microcalcifications (mcc) incorporating linear structure features in mammographic images by inputting a plurality of true mcc clusters and a plurality of normal clusters, gathering spot and cluster features from said clusters, extracting linear structure features, using said spot, cluster and linear structure features in mcc detector algorithm training, and applying said mcc detector algorithm to mammographic test images which are under testing for cancerous lesions to produce the result.

16. A computer readable storage for controlling a computer storing a program for detecting microcalcifications (mcc) incorporating linear structure features in mammographic images by inputting a plurality of true mcc clusters and a plurality of normal clusters, gathering spot and cluster features from said clusters, extracting linear structure features, using said spot, cluster and linear structure features in mcc detector algorithm training, and applying said mcc detector algorithm to mammographic test images which are under testing for cancerous lesions to produce the result.

17. A computer system configured for microcalcifications (mcc) detecting by forming a plurality of true mcc clusters and a plurality of normal clusters, gathering spot and cluster features from said clusters, extracting linear structure features, and using said spot, cluster and linear structure features in mcc detector algorithm training.

Assignments (8)