Automated method and system for computerized detection of masses and parenchymal distortions in medical images

• US 5,657,362 A
• Filed: 02/24/1995
• Issued: 08/12/1997
• Est. Priority Date: 02/24/1995
• Status: Expired due to Term

• Current Assignee: ARCH Development Corporation (University of Chicago)
• Original Assignee: ARCH Development Corporation (University of Chicago)
• Inventors: Zhang, Ming, Lu, Ping, Giger, Maryellen L.
• Primary Examiner(s): Porta, David P.
• Assistant Examiner(s): Bruce, David Vernon
• Application Number:
  US08/393,779
• Time in Patent Office: 900 Days
• Field of Search: 364/413.13, 364/413.23, 378/98.7, 378/37, 378/901, 382/130, 382/132, 382/171, 382/173
• US Class Current: 378/37
• CPC Class Codes: G06T 2207/10116 X-ray image
• CPC Class Codes: G06T 2207/20061 Hough transform
• CPC Class Codes: G06T 2207/20132 Image cropping
• CPC Class Codes: G06T 2207/30068 Mammography; Breast
• CPC Class Codes: G06T 2207/30096 Tumor; Lesion
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/11 Region-based segmentation

Abstract

A method and system for the automated detection of lesions such as masses and/or tissue (parenchymal) distortions in medical images such as mammograms. Dense regions and subcutaneous fat regions within a mammogram are segmented. A background correction may be performed within the dense regions. Hough spectrum within ROIs placed in the breast region of a mammogram are calculated and thresholded using the intensity value η in order to increase sensitivity and reduce the number of false-positive detections. Lesions are detected based on the thresholded Hough spectra. The thresholded Hough spectra are also used to differentiate between benign and malignant masses.

Claims

1. A method for enhancing visualization of a mammographic image, comprising:

identifying in said image an anatomically dense portion of said image; and

processing said dense portion to produce a processed image having a more uniform density.

2. A method as recited in claim 1, wherein identifying said dense portion comprises:

identifying subcutaneous fat in said image, said image having a plurality of pixels; and

excluding first pixels in said image corresponding to said subcutaneous fat to produce a fat-excluded image.

3. A method as recited in claim 2, comprising performing morphological filtering on said fat-excluded image.

4. A method as recited in claim 2, wherein excluding said first pixels comprises:

identifying a breast border in said image;

identifying a nipple position in said image;

selecting a first row of pixels including said nipple position and having a first starting pixel;

determining a gray-level threshold value in said image;

excluding second pixels in said first row having a gray-level value less than said threshold and being within a selected distance of said first starting pixel;

selecting a second row of pixels having a second starting pixel on said breast border; and

excluding third pixels in said second row having a gray-level value less than said threshold and being within said selected distance of said second starting pixel.

5. A method as recited in claim 4, wherein determining said threshold comprises:

placing a region on interest on said image proximate to a chest wall in said image;

generating a first gray-level histogram of fourth pixels in said region of interest;

generating a second gray-level histogram of said pixels in said image; and

determining said gray-level threshold based on said first and second gray-level histograms.

6. A method as recited in claim 1, wherein identifying said dense portion of said image comprises:

placing a region on interest on said image proximate to at least one of a chest wall and a breast border in said image, said image having a plurality of pixels;

generating a first gray-level histogram of first pixels in said region of interest;

generating a second gray-level histogram of said pixels in said image; and

determining a gray-level threshold based on said first and second gray-level histograms.

7. A method as recited in claim 6, comprising:

thresholding said image using said gray-level threshold to produce a thresholded image; and

performing morphological filtering on said thresholded image.

8. A method as recited in claim 6, comprising:

thresholding said image using said gray-level threshold to produce a thresholded image; and

performing at least one of feature analysis and multi-gray-level thresholding on said thresholded image.

9. A method as recited in claim 6, comprising:

thresholding said image using said gray-level threshold; and

calculating a percent dense of said image.

10. A method as recited in claim 9, wherein calculating said percent dense comprises:

determining a first number of pixels in said image above said gray-level threshold;

determining a second number of pixels within a breast region of said image excluding pixels corresponding to subcutaneous fat; and

dividing said first number by said second number.

11. A method as recited in claim 1, comprising:

background-trend correcting said dense portion.

12. A method as recited in claim 11, wherein background-trend correcting comprises:

curve fitting said dense portion to produce a fit to said dense portion;

subtracting said fit from said dense portion to produce a subtracted image; and

normalizing said subtracted image.

13. A method according to claim 1, comprising:

processing said first image to produce a second image;

extracting features from said second image; and

detecting a lesion in said image based on the extracted features extracted in said extracting step.

14. A method as recited in claim 13, wherein processing said dense portion is performed during extracting said features.

15. A method of detecting a lesion in a mammographic image, comprising:

determining a fat value of said image;

subtracting said fat value from said image to produce a subtracted image;

extracting features from said subtracted image; and

inputting said features to a detection scheme.

16. A method of detecting a lesion in a mammographic image, comprising:

obtaining right and left mammograms;

identifying in at least one of said right and left mammograms an anatomically dense portion;

processing said dense portion identified in said identifying step so that the mammogram in which the dense portion is identified has a more uniform density;

processing said right and left mammograms after said step of processing said dense portion to obtain a combined image;

performing feature analysis on said combined image to extract features from said combined image; and

detecting a lesion based on the extracted features.

17. A method as recited in claim 16, wherein said step of processing said dense portion comprises:

background-trend correcting said dense portion.

18. A method of analyzing a mammographic image, comprising:

selecting a plurality of regions of interest in said image;

calculating a Hough spectrum for each of said regions of interest;

determining a threshold using said Hough spectra;

thresholding said Hough spectra using said threshold;

identifying a region suspected of being a lesion in said image; and

analyzing said region.

19. A method as recited in claim 18, wherein determining said threshold comprises:

determining a mean and a first standard deviation of each of said Hough spectra;

determining an average mean and a second standard deviation over all of said regions of interest;

selecting a starting threshold;

determining if each of said Hough spectra are below said starting threshold;

varying said starting threshold to a next threshold and determining if each of said Hough spectra are below said next threshold; and

determining said threshold where a predetermined percentage of said Hough spectra are below said threshold.

20. A method as recited in claim 19, wherein selecting said starting threshold comprises:

selecting a starting value of η

;

determining said starting threshold as

space="preserve" listing-type="equation">τ

=μ

+η

σ

where μ

is said average mean and σ

is said second standard deviation.

21. A method as recited in claim 20, comprising:

incrementing η

to vary said starting threshold;

determining if η

is in an effective range; and

determining said threshold where said predetermined percentage of said Hough spectra are below said threshold when η

is no longer in said effective range.

22. A method as recited in claim 19, comprising:

detecting said lesion in an ROI having at least part of one of Hough spectra above said threshold.

23. A method as recited in claim 22, comprising:

determining a second threshold at which a second predetermined number of said Hough spectra are below said second threshold; and

detecting a parenchymal distortion in an ROI having at least part of one of said Hough spectra above said second threshold.

24. A method as recited in claim 22, comprising:

determining whether said lesion is malignant based on said accumulations; and

determining whether said lesion is benign based on said accumulations.

25. A method as recited in claim 19, comprising:

determining an accumulation for each of said Hough spectra thresholded at said threshold;

detecting said lesion based on said accumulation.

26. A method as recited in claim 25, comprising:

detecting a parenchymal distortion based on said accumulation.

27. A method as recited in claim 18, wherein said analyzing comprises:

measuring an accumulation for said region;

extracting at least one feature of said region using a feature analysis technique;

inputting said accumulation and said at least one feature into a lesion detection scheme; and

determining a likelihood of malignancy of said region using said lesion detection scheme.

28. A method as recited in claim 27, wherein inputting said accumulation comprises inputting said accumulation into an artificial neural network trained to determine malignancy of a suspect region in a mammogram.

29. A method as recited in claim 18, comprising:

performing preprocessing on said image prior to selecting said plurality of regions of interest.

30. A method as recited in claim 29, wherein said performing preprocessing comprises performing at least one dense portion processing, histogram equalization, peripheral enhancing and edge enhancing.

31. A method of analyzing an image, comprising:

locating a region suspected of containing a lesion;

selecting a plurality of regions of interest including said region;

calculating a Hough spectrum for each of said regions of interest;

determining a threshold using said Hough spectra;

thresholding said Hough spectra using said threshold to produce a thresholded Hough spectra;

determining an accumulation for each of said thresholded Hough spectra;

extracting at least one feature of said region using a feature analysis technique; and

analyzing said region using said accumulation and said at least one feature.

32. A method as recited in claim 31, wherein analyzing said region comprises:

inputting said accumulation and said at least one feature to a detection scheme; and

determining a likelihood of malignancy using said detection scheme.

33. A method as recited in claim 31, comprising:

obtaining a mammographic image;

detecting at least one of a mass and a parenchymal distortion in said mammographic image.

34. A method as recited in claim 31, comprising:

selecting a single region of interest containing said region;

calculating a single Hough spectrum for said single region of interest;

determining a threshold using said single Hough spectrum;

thresholding said single Hough spectrum using said threshold to produce a single thresholded Hough spectrum; and

determining an accumulation for said single thresholded Hough spectrum.

35. A method as recited in claim 34, wherein analyzing said region comprises:

inputting said accumulation and said at least one feature to a detection scheme; and

determining a likelihood of malignancy using said detection scheme.

36. A method as recited in claim 31, comprising:

detecting whether said region contains a malignant mass; and

detecting whether said region contains a benign mass.

37. A method as recited in claim 36, comprising differentiating between said malignant mass and said benign mass.

38. A system for visualization in an image, comprising:

an image acquisition device configured to acquire said image; and

a dense portion processing circuit connected to said image acquisition device and configured to identify an anatomically dense portion of said image and process said dense portion to produce a processed image having more uniform density.

39. A system as recited in claim 38, wherein said image acquisition device is a mammogram acquisition device.

40. The system of claim 38, comprising:

an image processing circuit connected to said dense portion processing circuit and configured to process said processed image having more uniform density;

a feature extraction circuit connected to said image processing circuit; and

a lesion detection circuit connected to said feature analysis circuit.

41. A system as recited in claim 40, comprising:

a lesion location superimposing circuit connected to said lesion detection circuit; and

an image display connected to said lesion location superimposing circuit.

42. A system for detecting lesions in an image;

an image acquisition device;

a lesion location circuit connected to said image acquisition device;

a dense portion processing circuit connected to said lesion location circuit;

a Hough spectrum analysis circuit connected to said lesion location circuit and said dense portion processing circuit;

a lesion extraction circuit connected to said Hough spectrum analysis circuit;

a feature extraction circuit connected to said lesion extraction circuit; and

a lesion detection circuit connected to said feature analysis circuit.

43. A system as recited in claim 42, comprising:

a lesion location superimposing circuit connected to said lesion detection circuit; and

an image display connected to said lesion location superimposing circuit.

44. A system as recited in claim 42, wherein said image acquisition device is a mammogram acquisition device.