Spiculated malignant mass detection and classification in a radiographic image

• US 8,958,625 B1
• Filed: 11/14/2014
• Issued: 02/17/2015
• Est. Priority Date: 04/30/2010
• Status: Active Grant

• Current Assignee: iCAD Incorporated
• Original Assignee: VuCOMP, Inc.
• Inventors: Wehnes, Jeffrey C., Harding, David S.
• Primary Examiner(s): Fujita, Katrina
• Application Number:
  US14/542,361
• Publication Number:
  US 20150071521A1
• Time in Patent Office: 95 Days
• Field of Search: 382/132.195
• US Class Current: 382/132
• 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 analysis embodiment comprises generating a bulge mask from a digital image, the bulge mask comprising potential convergence hubs for spiculated anomalies, detecting ridges in the digital image to generate a detected ridges map, projecting the detected ridges map onto a set of direction maps having different directional vectors to generate a set of ridge direction projection maps, determining wedge features for the potential convergence hubs from the set of ridge direction projection maps, selecting ridge convergence hubs from the potential convergence hubs having strongest wedge features, extracting classification features for each of the selected ridge convergence hubs, and classifying the selected ridge convergence hubs based on the extracted classification features.

Claims

1. An anomaly detection system for identifying spiculated anomalies in an image comprising pixels, the system comprising:

a processor; and

a non-transitory computer readable storage medium storing programming for execution by the processor, the programming including instructions for;

generating a bulge mask from a digital image, the bulge mask comprising potential convergence hubs for spiculated anomalies;

detecting ridges in the digital image to generate a detected ridges map;

projecting the detected ridges map onto a set of direction maps having different directional vectors to generate a set of ridge direction projection maps, wherein projecting the detected ridges map onto the set of direction maps comprises;

separating the detected ridges map into a line mask image, a row component image, and a column component image; and

determining a dot product of the line mask image, the row component image, and the column component image with the directional vectors;

determining wedge features for the potential convergence hubs from the set of ridge direction projection maps;

selecting ridge convergence hubs from the potential convergence hubs having strongest wedge features;

extracting classification features for each of the selected ridge convergence hubs; and

classifying the selected ridge convergence hubs based on the extracted classification features.

2. The anomaly detection system of claim 1, wherein the image is a mammogram, the bulges are potentially-malignant masses, the spiculated anomalies are potentially-malignant spiculated masses, and the one or more classification features are selected from the group consisting of:

wedge width, radius, signal-to-noise ratio (SNR), minimum hub contrast, nipple distance and y position, and combinations thereof.

3. The anomaly detection system of claim 1, wherein the instructions for detecting the ridges comprise instructions for:

detecting the ridges at multiple ridge widths, multiple ridge orientations, and multiple ridge aspect ratios;

determining an image noise map for each ridge width;

calculating second derivative measurements at each ridge width, orientation and aspect ratio;

determining an average contrast and contrast standard deviation from the second derivative measurements;

determining the SNR from the average contrast and the noise map; and

selecting the ridges to include in the detected ridges map based on relative values of at least one of SNR, contrast density, and contrast standard deviation.

4. The anomaly detection system of claim 3, wherein programming further includes instructions for thinning the selected ridges.

5. The anomaly detection system of claim 1, wherein the instructions for determining the wedge features comprise instructions for:

measuring the wedge features for multiple wedge widths, multiple wedge offsets, and multiple wedge radii; and

generating a set of radius/wedge width maps comprising a highest SNR and respective offset for each potential convergence hub.

6. The anomaly detection system of claim 5, wherein the instructions for selecting ridge convergence hubs comprise instructions for:

thresholding each radius/wedge width map to a minimum significant SNR; and

thinning contiguous groups of ridge convergence hubs to one ridge convergence hub having maximum relative SNR.

7. The anomaly detection system of claim 1, wherein the instructions for detecting the ridges comprise instructions for detecting the ridges for multiple ridge widths, multiple ridge orientations, and multiple ridge aspect ratios.

8. The anomaly detection system of claim 1, further comprising instructions for removing bright areas from and flattening an intensity of the digital image, before generating the bulge mask.

9. The anomaly detection system of claim 1, further comprising instructions for marking, on an output image, ones of the selected ridge convergence hubs having extracted classification features exceeding a threshold, after classifying the selected ridge convergence hubs.

10. An anomaly detection system for identifying spiculated anomalies in an image comprising pixels, the system comprising:

a non-transitory computer readable storage medium storing a digital image;

a processor coupled to the memory and configured for;

generating a bulge mask from the digital image, the bulge mask comprising potential convergence hubs for spiculated anomalies;

detecting ridges in the digital image to generate a detected ridges map;

projecting the detected ridges map onto a set of direction maps having different directional vectors to generate a set of ridge direction projection maps, wherein projecting the detected ridges map onto the set of direction maps comprises;

separating the detected ridges map into a line mask image, a row component image, and a column component image; and

determining a dot product of the line mask image, the row component image, and the column component image with the directional vectors;

determining wedge features for the potential convergence hubs from the set of ridge direction projection maps;

selecting ridge convergence hubs from the potential convergence hubs having strongest wedge features;

extracting classification features for each of the selected ridge convergence hubs;

classifying the selected ridge convergence hubs based on the extracted classification features;

generating an output image in accordance with the classified selected ridge convergence hubs; and

saving the output image to the non-transitory computer readable storage medium.

11. The anomaly detection system of claim 10, wherein the image is a mammogram, the bulges are potentially-malignant masses, the spiculated anomalies are potentially-malignant spiculated masses, and the one or more classification features are selected from the group consisting of:

wedge width, radius, signal-to-noise ratio (SNR), minimum hub contrast, nipple distance and y position, and combinations thereof.

12. The anomaly detection system of claim 10, wherein processor configured for detecting the ridges comprise the processor configured for:

detecting the ridges at multiple ridge widths, multiple ridge orientations, and multiple ridge aspect ratios;

determining an image noise map for each ridge width;

calculating second derivative measurements at each ridge width, orientation and aspect ratio;

determining an average contrast and contrast standard deviation from the second derivative measurements;

determining the SNR from the average contrast and the noise map; and

selecting the ridges to include in the detected ridges map based on relative values of at least one of SNR, contrast density, and contrast standard deviation.

13. The anomaly detection system of claim 12, wherein the processor is further configured for thinning the selected ridges.

14. The anomaly detection system of claim 10, wherein the processor configured for determining the wedge features comprise the processor configured for:

measuring the wedge features for multiple wedge widths, multiple wedge offsets, and multiple wedge radii; and

generating a set of radius/wedge width maps comprising a highest SNR and respective offset for each potential convergence hub.

15. The anomaly detection system of claim 14, wherein the processor configured for selecting ridge convergence hubs comprise instructions for:

thresholding each radius/wedge width map to a minimum significant SNR; and

thinning contiguous groups of ridge convergence hubs to one ridge convergence hub having maximum relative SNR.

16. The anomaly detection system of claim 10, wherein the processor configured for detecting the ridges comprise the processor configured for detecting the ridges for multiple ridge widths, multiple ridge orientations, and multiple ridge aspect ratios.

17. The anomaly detection system of claim 10, further comprising the processor configured for removing bright areas from and flattening an intensity of the digital image, before generating the bulge mask.

18. The anomaly detection system of claim 10, wherein the processor configured for generating the output image comprises the processor configured for marking, on the output image, ones of the selected ridge convergence hubs having extracted classification features exceeding a threshold, after classifying the selected ridge convergence hubs.