Method and apparatus for automatic muscle segmentation in digital mammograms

• US 6,035,056 A
• Filed: 03/27/1997
• Issued: 03/07/2000
• Est. Priority Date: 03/27/1997
• Status: Expired due to Fees

• Current Assignee: Hologic Incorporated
• Original Assignee: R2 Technology, Inc. (Hologic Incorporated)
• Inventors: Karssemeijer, Nico
• Primary Examiner(s): Chang, Jon
• Assistant Examiner(s): Patel, Jayanti K.
• Application Number:
  US08/825,291
• Time in Patent Office: 1,076 Days
• Field of Search: 128/922, 382/128, 382/132, 382/171, 382/199, 382/281, 382/130, 382/131, 382/308, 378/37
• US Class Current: 382/132
• CPC Class Codes: G06T 2207/10116 X-ray image
• CPC Class Codes: G06T 2207/20061 Hough transform
• 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 7/12 Edge-based segmentation
• CPC Class Codes: G06V 10/48 by mapping characteristic values of the pattern into a parameter space, e.g. Hough transformation

Abstract

A method and apparatus for the automatic segmentation of the pectoral muscle boundary in a digital mammogram is disclosed. Parameters of the pectoral boundary line are determined by calculating gradient magnitudes of the digital mammogram in a region of interest, accumulating the gradient magnitudes into a parameter plane according to a Hough transform, and using information in the parameter plane for identifying the pectoral boundary. Prior to being accumulated in the parameter plane, the gradient magnitudes are weighted by a monotonic, multi-bit weighting function designed to exhibit larger variations near more densely populated levels of gradient magnitudes, and designed to exhibit smaller variations near less densely populated levels of gradient magnitudes, for allowing greater robustness against variations in gradient magnitudes among different digital mammograms.

Claims

1. A method of identifying a pectoral boundary in a digital mammogram, said pectoral boundary being characterized by at least one parameter, comprising the steps of:

determining a region of interest containing said pectoral boundary;

calculating gradient magnitudes inside said region of interest;

accumulating said gradient magnitudes into a parameter plane, represented by H(ρ

,θ

), according to a Hough transform; and

using information in said parameter plane for identifying said at least one parameter of said pectoral boundary;

wherein said step of using information in said parameter plane comprises the steps of;

selecting a parameter plane region of interest corresponding to the range of possible angles normal to said pectoral boundary;

creating a normalized parameter plane HN(ρ

,θ

) by normalizing the value H(ρ

,θ

) by a normalizing factor, said normalizing factor being inversely related to the number of gradient magnitude pixels capable of being accumulated at (ρ

,θ

);

identifying a highest ranking local peak of said normalized parameter plane HN(ρ

,θ

); and

using the location of said highest ranking local peak for determining said at least one parameter of said pectoral boundary; and

wherein said normalizing factor is determined for each pixel (ρ

,θ

) by the steps of;

backprojecting a line corresponding to (ρ

,θ

) into the image plane;

determining the length L of said line; and

setting said normalizing factor equal to 1/sqrt(L) if L is greater than a first number of pixels.

2. The method of claim 1, wherein said step of accumulating said gradient magnitudes comprises the step of weighting said gradient magnitudes according to a monotonic, multi-bit weighting function designed to exhibit larger variations near more densely populated levels of gradient magnitudes, and designed to exhibit smaller variations near less densely populated levels of gradient magnitudes.

3. The method of claim 2, wherein said weighting function for a selected gradient magnitude is proportional to the integral of a histogram of said gradient magnitudes taken between a first gradient magnitude and said selected gradient magnitude.

4. The method of claim 3, said digital mammogram being represented by M(x,y), said gradient magnitudes being represented by g(x,y), said gradient angles being represented by fi(x,y), said parameter plane being represented by H(ρ

,θ

), and said weighting function being represented by W(g), said step of accumulating said gradient magnitudes into a parameter plane using a Hough transform comprising the steps of;

for each gradient magnitude at pixel (x,y) being accumulated, determining a set of parameter plane destination pixels (ρ

,θ

) to be accumulated; and

incrementing H(ρ

,θ

) at each of said parameter plane destination pixels by an amount equal to W(g(x,y)).

5. The method of claim 4, wherein for each gradient plane pixel at (x,y), said set of parameter plane destination pixels is selected according to the constraint ρ

=x cos (θ

)+y sin (θ

), and wherein only those parameter plane destinations having θ

within a fixed interval Δ

θ

_INTERVAL of fi(x,y) are incremented.

6. The method of claim 1, said step of identifying a highest ranking local peak further comprising the steps of:

determining whether a first set of peaks of HN(ρ

, θ

) above a lower predetermined threshold T_L can be found;

if such a first set of peaks cannot be found, selecting a NULL value as the highest ranking local peak;

if such a first set can be found, determining if a second set of peak of HN(ρ

,θ

) above an upper predetermined threshold T_H can be found;

if such a second set of peaks can be found, determining the corresponding area A of a segmented pectoral muscle in image space corresponding to each of said second set of peaks, and selecting that peak having the greatest corresponding area A as the highest ranking local peak;

if such a second set of peaks above T_H cannot be found, selecting the largest of said first set of peaks as the highest ranking local peak.

7. The method of claim 1, said gradient magnitude plane being N pixels on a side, said first number of pixels being approximately N/10, wherein said normalizing factor set to 1/sqrt(N/10) if L is less than approximately N/10 pixels.

8. The method of claim 1, further comprising the step of computing gradient angles inside said region of interest, wherein said step of accumulating gradient magnitudes comprises the step of only accumulating gradient magnitudes associated with a predetermined range of gradient angles.

9. The method of claim 8, said pectoral boundary defining a pectoral muscle portion of the digital mammogram, said pectoral muscle portion encompassing a first corner of said digital mammogram, wherein said predetermined range of gradient angles includes an angle corresponding to said first corner and a first range of angles therearound.

10. The method of claim 9, wherein said first range of angles is approximately 90 degrees.

11. The method of claim 10, said digital mammogram comprising a tissue portion including a pectoral muscle portion and a breast tissue portion, said pectoral boundary defining said pectoral muscle portion, said pectoral muscle portion encompassing a first corner of said digital mammogram, said step of determining a region of interest including the steps of:

determining a center of mass of said tissue area; and

selecting said region of interest above a line centered at said center of mass and having a predetermined slope.

12. The method of claim 11, wherein said predetermined slope is approximately 1.5.

13. A method of identifying a boundary between a first portion and a second portion of a digitized radiograph, said second portion corresponding to a tissue area of generally higher density than said first portion, said boundary generally being a linear boundary, comprising the steps of:

determining a region of interest containing said boundary;

calculating gradient magnitudes inside said region of interest;

accumulating said gradient magnitudes into a parameter plane, represented by H(ρ

,θ

), according to a transform function;

using information in said parameter plane for identifying said linear boundary;

wherein said step of using information in said parameter plane comprises the steps of;

selecting a parameter plane region of interest corresponding to the range of possible angles normal to said pectoral boundary;

creating a normalized parameter plane HN(ρ

,θ

) by normalizing the value H(ρ

,θ

) by a normalizing factor, said normalizing factor being inversely related to the number of gradient magnitude pixels capable of being accumulated at (ρ

,θ

);

identifying a highest ranking local peak of said normalized parameter plane HN(ρ

,θ

); and

using the location of said highest ranking local peak for determining said at least one parameter of said linear boundary; and

wherein said normalizing factor is determined for each pixel (ρ

,θ

) by the steps of;

backprojecting a line corresponding to (ρ

,θ

) into the image plane;

determining the length L of said line;

setting said normalizing factor equal to 1/sqrt(L) if L is greater than a first number of pixels.

14. The method of claim 13, wherein said step of accumulating said gradient magnitudes comprises the step of weighting said gradient magnitudes according to a monotonic, multi-bit weighting function designed to exhibit larger variations near more densely populated levels of gradient magnitudes, and designed to exhibit smaller variations near less densely populated levels of gradient magnitudes.

15. The method of claim 13, wherein said transform function is a Hough transform, and wherein for each gradient plane pixel at (x,y), said set of parameter plane destination pixels is selected according to the constraint ρ

=x cos (θ

)+y sin (θ

), and wherein only those parameter plane destinations having θ

within a fixed interval Δ

θ

_INTERVAL of fi(x,y) are incremented.

16. The method of claim 15, said highest ranking local peak being at a location (ρ

1,θ

1), said linear boundary being characterized by an angle parameter and an offset parameter, wherein said angle parameter is set equal to θ

1 and said offset parameter is set equal to ρ

1.

17. A method of identifying a pectoral boundary in a digital mammogram D(x,y), said pectoral boundary being a linear boundary characterized by an offset parameter and an angle parameter, comprising the steps of:

determining a region of interest containing said pectoral boundary;

calculating gradient magnitudes g(x,y) inside said region of interest;

weighting said gradient magnitudes g(x,y) by a monotonic, multi-bit weighting function designed to exhibit larger variations near more densely populated levels of gradient magnitudes, and designed to exhibit smaller variations near less densely populated levels of gradient magnitudes, for producing weighted gradient magnitudes;

accumulating said weighted gradient magnitudes into a parameter plane H(ρ

,θ

) according to a Hough transform ρ

=x cos (θ

)+y sin (θ

);

creating a normalized parameter plane HN(ρ

,θ

) by normalizing the value H(ρ

,θ

) by a normalizing factor, said normalizing factor being inversely related to the number of gradient magnitude pixels capable of being accumulated at (ρ

,θ

);

identifying a highest ranking local peak of said normalized parameter plane HN(ρ

,θ

), said highest ranking local peak being at a location (ρ

_P,θ

_P), wherein said angle parameter is set equal to θ

_P and said offset parameter is set equal to ρ

_P ;

wherein said digital mammogram comprises a tissue portion including a pectoral muscle portion and a breast tissue portion, said pectoral boundary defining said pectoral muscle portion, said pectoral muscle portion encompassing a first corner of said digital mammogram, said step of determining a region of interest including the steps of;

determining a center of mass of said tissue area; and

selecting said region of interest above a line centered at said center of mass and having a predetermined slope.

18. A method of identifying a pectoral boundary in a digital mammogram, said pectoral boundary being characterized by at least one parameter, comprising the steps of:

determining a region of interest containing said pectoral boundary;

calculating gradient magnitudes inside said region of interest;

computing gradient angles inside said region of interest;

accumulating only gradient magnitudes associated with a predetermined range of angles into a parameter plane according to a Hough transform; and

using information in said parameter plane for identifying said at least one parameter of said pectoral boundary;

wherein said digital mammogram comprises a tissue portion including a pectoral muscle portion and a breast tissue portion, said pectoral boundary defines a pectoral muscle portion of the digital mammogram, said pectoral muscle portion encompasses a first corner of said digital mammogram, wherein said predetermined range of gradient angles includes an angle corresponding to said first corner and a first range of angles that is approximately 90 degrees and said step of determining a region of interest includes the steps of;

determining a center of mass of said tissue area; and

selecting said region of interest above a line centered at said center of mass and having a predetermined slope.

19. A method of identifying a pectoral boundary in a digital mammogram, said pectoral boundary being characterized by at least one parameter, comprising the steps of:

determining a region of interest containing said pectoral boundary;

calculating gradient magnitudes inside said region of interest;

accumulating said gradient magnitudes into a parameter plane, represented by H(ρ

,θ

), according to a Hough transform; and

using information in said parameter plane for identifying said at least one parameter of said pectoral boundary;

wherein said step of using information in said parameter plane comprises the steps of;

selecting a parameter plane region of interest corresponding to the range of possible angles normal to said pectoral boundary;

said parameter plane being represented by H(ρ

,θ

), wherein said step of using information in said parameter plane comprises the steps of;

selecting a parameter plane region of interest corresponding to the range of possible angles normal to said pectoral boundary;

creating a normalized parameter plane HN(ρ

, θ

) by normalizing the value H(ρ

,θ

) by a normalizing factor, said normalizing factor being inversely related to the number of gradient magnitude pixels capable of being accumulated at (ρ

,θ

);

identifying a highest ranking local peak of said normalized parameter plane HN(ρ

,θ

);

using the location of said highest ranking local peak for determining said at least one parameter of said pectoral boundary; and

said step of identifying a highest ranking local peak further comprises the steps of;

determining whether a first set of peaks of HN(ρ

, θ

) above a lower predetermined threshold T_L can be found;

if such a first set of peaks cannot be found, selecting a NULL value as the highest ranking local peak;

if such a first set can be found, determining if a second set of peaks of HN(ρ

,θ

) above an upper predetermined threshold T_H can be found;

if such a second set of peaks can be found, determining the corresponding area A of a segmented pectoral muscle in image space corresponding to each of said second set of peaks, and selecting that peak having the greatest corresponding area A as the highest ranking local peak;

if such a second set of peaks above T_H cannot be found, selecting the largest of said first set of peaks as the highest ranking local peak.