Method for computer-aided detection of clustered microcalcifications from digital mammograms

• US 5,537,485 A
• Filed: 07/21/1992
• Issued: 07/16/1996
• Est. Priority Date: 07/21/1992
• Status: Expired due to Fees

• Current Assignee: ARCH Development Corporation (University of Chicago)
• Original Assignee: ARCH Development Corporation (University of Chicago)
• Inventors: Giger, Maryellen L., Nishikawa, Robert M., Doi, Kunio
• Primary Examiner(s): Harvey, Jack B.
• Assistant Examiner(s): CHUNG TRANS, XUONG MY
• Application Number:
  US07/915,631
• Time in Patent Office: 1,456 Days
• Field of Search: 364/413.13, 364/413.22, 382/6, 382/19, 382/130, 382/132, 382/133, 382/128, 382/170, 382/172, 382/169, 382/257, 128/664
• US Class Current: 382/130
• CPC Class Codes: G06T 2207/30068 Mammography; Breast
• CPC Class Codes: G06T 7/0012 Biomedical image inspection

Abstract

A method of computerized detection of clustered microcalcifications in digital mammograms, including obtaining a digitized mammogram, deriving a difference image signal from the digitized mammogram, performing global grey-level thresholding, area filtering, and local grey-level thresholding on the difference image, in that order, performing a texture discrimination of the signal extracted from the previous step, performing a cluster filtering technique on the texture discriminated signals to identify locations in the digitized mammogram corresponding to candidate clustered microcalcifications, performing a feature extraction step in which the area, contrast and background pixel values of signals corresponding to the candidate clustered microcalcifications in the original image are extracted, performing thresholding tests based on the extracted features to eliminate from the candidate clustered microcalcifications particular candidate clustered microcalcification identified as corresponding to false-positive identifications, cluster filtering the remaining candidate clustered microcalcifications to eliminate further candidate clustered microcalcifications which are not sufficiently clustered, and outputting to a radiologist an image indicating, by use of arrows, the positions of the remaining clustered microcalcifications.

Claims

1. A method of computerized detection of clustered microcalcifications in a digital mammogram derived from an original radiographic image, comprising:

producing a multi-pixel difference image of said digital mammogram to remove low-frequency structured anatomic background;

performing a global gray-level thresholding of said difference image to produce a global thresholded image, including

producing a histogram of the difference image,

setting a global threshold based on a pre-selected percentage of pixels at the high end of the histogram,

retaining unchanged in said difference image those pixels having a value above said global threshold, and

setting to a predetermined constant the values of those pixels in the difference image which have a value below said global threshold;

performing area filtering by means of morphological erosion on the image data resulting after global thresholding thereby to produce an area filtered image in which small signals are selectively removed; and

processing said area filtered image in which small signals are selectively removed to identify microcalcifications;

wherein said step of performing area filtering comprises;

defining plural structuring elements each having a size of at least two pixels,

sequentially defining each pixel in the image after global thresholding as an operating pixel and sequentially overlaying each operating pixel with each structuring element;

determining, for each operating pixel sequentially overlaid by each structuring element, the minimum pixel value of the pixels overlaid by each structuring element in relation to the same operating pixel;

determining the maximum of the minimum pixel values determined in relation to the same operating pixel; and

setting the value of each pixel in the output area-filtered image corresponding to the position of the operating pixel in the global thresholded image to the respective maximum of said minimum values determined in the preceding step.

2. The method according to claim 1, wherein said step of processing said area filtered image comprises:

performing local gray-level thresholding on the area filtered image to produce a threshold image in which pixels having a value below a local threshold are set to a uniform background value and pixels having values above the local threshold retain their values;

performing texture discrimination on the threshold image by removing background trend therefrom and setting to a predetermined value those pixels having a first moment of power spectrum greater than a predetermined value thereby to produce a texture discriminated image;

performing cluster filtering of the texture discriminated image to produce a cluster image of single-pixel signals identifying locations of potential microcalcifications;

extracting predetermined features from locations in the digital mammogram identified in said cluster filtering step; and

identifying microcalcifications based on the features extracted in said extracting step.

3. The method according to claim 2, wherein said cluster filtering step comprises:

a step of area-point transformation in which signals representing possible microcalcifications are reduced to respective single pixels in a first binary image;

a step of counting the number of pixels exceeding a predetermined value in a kernel of a predetermined size scanned across said first binary image and copying the contents of the said first binary image that lay within the kernel to the corresponding locations in a second binary image whenever the contents of the kernel meet a predetermined grouping criteria based on said counting;

outputting a second binary image comprised of the pixels remaining after the counting and eliminating step.

4. The method according to claim 3, wherein said cluster filtering step comprising:

selecting the size of said kernel so as to discriminate against false clusters represented by pixels in the first binary image separated by greater than a predetermined distance.

5. The method according to claim 2, wherein said step of extracting predetermined features comprises:

determining, for each region in the digital mammogram of predetermined size corresponding to the location of each pixel of the second binary image, at least two of the area, contrast and background value of a signal within each said region.

6. The method according to claim 5, wherein said identifying step comprises:

performing an area versus contrast thresholding test to determine, for each signal for which features are extracted, whether that signal corresponds to a false positive signal, and if so, removing from said second binary image that pixel identifying the location of said false positive signal by setting the value of that pixel in the second binary image to a predetermined value.

7. This method according to claim 6, wherein said identifying step comprises:

performing a contrast versus background test to determine for each signal for which features are extracted, whether that signal corresponds to a false positive signal, and if so, removing from said second binary image that pixel identifying the location of said false positive signal by setting the value of that pixel in the second binary image to a predetermined value.

8. The method according to claim 7, wherein said identifying step further comprises:

a step of counting the number of pixels exceeding a predetermined value in a kernel of a predetermined size scanned across the binary image remaining after said area versus contrast thresholding test and said contrast versus background thresholding test, and copying the contents of the said binary image that lay within the kernel to the corresponding locations in a third binary image whenever the contents of the kernel meet a predetermined grouping criteria based on said counting; and

indicating locations of microcalcifications in said original radiographic image based on the locations of pixels having a predetermined value in said third binary image.

9. The method according to claim 6, wherein said identifying step further comprises:

a step of counting the number of pixels exceeding a predetermined value in a kernel of a predetermined size scanned across the binary image remaining after said area versus contrast thresholding test, and copying the contents of the said binary image that lay within the kernel to the corresponding locations in a third binary image whenever the contents of the kernel meet a predetermined grouping criteria based on said counting; and

indicating locations of microcalcifications in said original radiographic image based on the locations of pixels having a predetermined value in said third binary image.

10. This method according to claim 5, wherein said identifying step comprises:

performing a contrast versus background test to determine for each signal for which features are extracted, whether that signal corresponds to a false positive signal, and if so, removing from said second binary image that pixel identifying the location of said false positive signal by setting the value of that pixel in the second binary image to a predetermined value.

11. The method according to claim 10, wherein said identifying step further comprises;

a step of counting the number of pixels exceeding a predetermined value in a kernel of a predetermined size scanned across the binary image remaining after said contrast versus background thresholding test, and copying the contents of the said binary image that lay within the kernel to the corresponding locations in a third binary image whenever the contents of the kernel meet a predetermined grouping criteria based on said counting; and

indicating locations of microcalcifications in said original radiographic image based on the locations of pixels having a predetermined value in said third binary image.

12. The method according to claim 2, wherein said step of performing local gray-level thresholding on the area filtered image comprises:

determining said local threshold at least in part based on a preselected multiple of the standard deviation of the pixel value in a region surrounding the respective pixels in said area filtered image.

13. The method of claim 1, wherein said step of defining plural structuring elements comprises:

defining said structuring elements to include at least one subset of all possible combinations of two interconnected pixels.

14. The method of claim 1, wherein said step of defining plural structuring elements comprises:

defining said structuring elements to include at least one subset of all possible combinations of three interconnected pixels.

15. A method of computerized detection of clustered microcalcifications in a digital mammogram derived from an original radiographic image, comprising:

filtering said digital mammogram to produce a filtered image from which signals meeting predetermined criteria are removed;

performing a first cluster filtering on said filtered image to produce a cluster image of single-pixel signals identifying locations of prospective problematic microcalcifications;

extracting predetermined features from signals at locations in the digital mammogram identified by the single-pixel signals in said cluster image;

performing at least one predetermined thresholding test based on the extracted features and based on said predetermined thresholding test, determining whether the signals from which the features were extracted correspond to a false-positive microcalcification cluster;

eliminating from said cluster image those single-pixel signals corresponding to locations of false-positive microcalcification clusters;

performing a second cluster filtering on the cluster image remaining after said eliminating step and outputting a further cluster image comprised of single-pixel values; and

indicating in said original radiographic image locations of clustered microcalcifications based on the locations of pixels having a predetermined value in said further cluster image.

16. The method according to claim 15, wherein said filtering step comprises:

producing a multi-pixel difference image of said digital mammogram to remove low-frequency structured anatomic background;

performing a global grey-level thresholding of said difference image, including

producing a histogram of the difference image,

setting a global threshold based on a pre-selected percentage of pixels at the high end of the histogram,

retaining unchanged in said difference image those pixels having a value above said global threshold, and

setting to a predetermined constant the values of those pixels in the difference image which have a value below said global threshold;

performing area filtering by means of morphological erosion on the image data resulting after global thresholding thereby to produce an area filtered image in which small signals are selectively removed;

performing local grey-level thresholding on the area filtered image to produce a threshold image in which pixels having a value below a local threshold determined at least in part based on a pre-selected multiple of the standard deviation of the pixel value in a region surrounding the respective pixels in said area filtered image are set to a uniform background value and pixels having values above the local threshold retain their values; and

performing texture discrimination on the threshold image by removing background trend therefrom and setting to a predetermined value those pixels having a first moment of power spectrum greater than a predetermined value thereby to produce a texture discriminated image.

17. The method according to claim 16, wherein said step of area filtering comprises:

defining plural structuring elements each having a size of at least two pixels;

sequentially defining each pixel in the image after global thresholding as an operating pixel and sequentially overlaying each operating pixel with each structuring element;

determining, for each operating pixel sequentially overlaid by each structuring element, the minimum pixel value of the pixels overlaid by each structuring element in relation to the same operating pixel;

determining the maximum of the minimum pixel values determined in relation to the same operating pixel; and

setting the value of each pixel in the output area-filtered image corresponding to the position of the operating pixel in the global thresholded image to the respective maximum of said minimum values determined in the preceding step.

18. The method of claim 17, wherein sad step of defining plural structuring elements comprises:

defining said structuring elements to include at least one subset of all possible combinations of two interconnected pixels.

19. The method of claim 17, wherein said step of defining plural structuring elements comprises:

defining said structuring elements to include at least one subset of all possible combinations of two interconnected pixels.

20. The method according to claim 15, wherein said first cluster filtering step comprises:

a step of area-point transformation in which signals representing possible microcalcifications are reduced to respective single pixels in a first binary image;

a step of counting the number of pixels exceeding a predetermined value in a kernel of predetermined size scanned across said first binary image and copying the contents of the said binary image that lay within the kernel to the corresponding locations in a second binary image whenever the contents of the kernel meet a predetermined grouping criteria based on said counting; and

outputting said second binary image comprised of the pixels remaining after the counting and copying step.

21. The method according to claim 20, wherein said second cluster filtering step comprises:

a step of counting the number of pixels exceeding a predetermined value in a kernel of predetermined size scanned across the cluster image remaining after said eliminating step, and copying the contents of said cluster image that lay within the kernel to the corresponding locations in a third binary image whenever the contents of the kernel meet a predetermined grouping criteria based on said counting.

22. The method according to claim 20, wherein said step of extracting predetermined features comprises:

determining, for each region in said digital mammogram of predetermined size corresponding to the location of each pixel in said cluster image, at least two of area, contrast and background values of a signal within each said region.

23. The method according to claim 22, wherein said step of performing at least one predetermined thresholding test comprises:

performing an area versus contrast thresholding test to determine, for each signal for which features are extracted, whether that signal corresponds to a false positive microcalcification cluster, and if so, removing from said cluster image that pixel identifying the location of said false positive microcalcification cluster by setting the value of that pixel in the cluster image to a predetermined value.

24. This method according to claim 23, wherein said of performing at least one predetermined thresholding test identifying comprises:

performing a contrast versus background test to determine for each signal for which features are extracted, whether that signal corresponds to a false positive signal, and if so, removing from said cluster image that pixel identifying the location of said false positive signal by setting the value of that pixel in the cluster image to a predetermined value.

25. This method according to claim 22, wherein said step of performing at least one predetermined thresholding test comprises:

performing a contrast versus background test to determine for each signal for which features are extracted, whether that signal corresponds to a false positive signal, and if so, removing from said cluster image that pixel identifying the location of said false positive signal by setting the value of that pixel in the cluster image to a predetermined value.

26. The method according to claim 20, wherein said first cluster filtering step comprises:

selecting the size of said kernel so as to discriminate against false clusters represented by pixels in the first binary image separated by greater than a predetermined distance.