METHODS AND APPARATUSES FOR CORRECTING A MAMMOGRAM WITH AN IMPLANT, AND FOR SEGMENTING AN IMPLANT

• US 20080002872A1
• Filed: 06/28/2007
• Published: 01/03/2008
• Est. Priority Date: 06/30/2006
• Status: Active Grant

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Durand, Jerome, GATESOUPE, Pascal, Razvan, Iordache, Lavoue, Sophie, Muller, Serge
• Granted Patent: US 8,160,346 B2
• Time in Patent Office: Days
• US Class Current: 382/132
• CPC Class Codes: A61B 6/502 for diagnosis of breast, i.e. mammography
• CPC Class Codes: A61B 6/5252 removing objects from field of view, e.g. removing patient table from a CT image
• CPC Class Codes: G06T 2207/10116 X-ray image
• CPC Class Codes: G06T 2207/30052 Implant; Prosthesis
• CPC Class Codes: G06T 2207/30068 Mammography; Breast
• CPC Class Codes: G06T 2207/30096 Tumor; Lesion
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06T 7/155 involving morphological operators

Abstract

The invention relates to a method of correcting a digital mammogram of a breast containing an implant. According to the invention, brightnesses situated at and around the implant peak (PI) in the low frequency component of the image (X_LP) are attenuated (ATT) as compared with a function (P1) that is applied to the other brightnesses, thereby attenuating the implant.

The invention also provides a method of segmenting an implant in a radiographic image, the method being characterized by the following steps: smoothing the image by a noise-reducing lowpass filter, applying a non-directional spatial derivative filter, pixel thresholding, and selecting an outline to define the segmented portion of the image that corresponds to the implant.

Claims

1. A method of segmenting an implant in a radiographic image of a portion of a patient'"'"'s body that contains said implant, to enable the segmented portion of the image corresponding to the implant to be processed subsequently, the method comprising the following steps performed by a computer:

smoothing (S5) the image by a lowpass first filter for reducing noise, in order to obtain a smoothed image;

applying (S6) a non-directional spatial derivative second filter to the smoothed image, in order to obtain a filtered image;

thresholding (S7) pixels of the filtered image in order to obtain a thresholded image, the thresholding (S7) of the filtered image is hysteresis thresholding, comprising a high first threshold and a low second threshold, the thresholding (S7) of the filtered image comprising retaining pixels having a brightness value greater than or equal to the high first threshold, discarding pixels having a brightness value less than the low second threshold, and retaining pixels having a brightness value lying between the high first threshold and the low second threshold providing they are adjacent to at least one pixel having a brightness value greater than or equal to the high first threshold;

selecting (S8) at least one outline obtained in the thresholded image, in order to define the segmented portion of the image corresponding to the implant;

modifying (S12) the threshold(s) used during the thresholding step (S7) if the segmented portion that is obtained does not satisfy at least one calculated criterion, and reapplying the thresholding and selection steps (S7, S8) in order to obtain a new segmented portion,

said threshold modification (S12) includes at least modifying the low second threshold; and

if said criterion is not satisfied after a predetermined number of modifications or after lowering the low second threshold down to a minimum, then dilatations are performed followed by erosions of increasing size on the thresholded image until a closed outline is obtained for defining the segmented portion.

2. A method of segmenting an implant according to claim 1, wherein the image smoothing (S5) is performed by a Gaussian lowpass first filter.

3. A segmentation method according to claim 1, wherein the non-directional spatial derivative second filter (S6) applied to the smoothed image is a Laplacian filter.

4. A method of segmenting an implant according to claim 1, wherein the image smoothing (S5) is performed by a Gaussian lowpass first filter, and the non-directional spatial derivative second filter (S6) applied to the smoothed image is a Laplacian filter, the Gaussian first filter and the Laplacian second filter together forming a global LoG filter implementing the following function:

5. A segmentation method according to claim 1, wherein zero-crossings are looked for in the filtered image, as subjected to the gradient norm calculated on said filtered image, in order to preselect at least one outline.

6. A method of segmenting an implant according to claim 1, wherein the criterion is at least one of the group constituted by:

the area of the segmented portion belongs to a range of areas relative to the estimated area of the portion of the body in the image; and

the variance of the segmented portion is below a variance threshold.

7. A segmentation method according to claim 1, wherein the step (S8) of selecting at least one outline obtained in the thresholded image is performed automatically by comparison with region models defined by prerecorded reference functions.

8. A segmentation method according to claim 1, wherein processing (S9) is applied to the resulting segmented image portion, which processing consists in applying a mask for blocking out said segmented portion from the initial image.

9. A method of segmenting an implant according to claim 1, wherein processing is applied to the resulting segmented portion of the image, which processing consists in identifying its outline in the initial image.

10. A method of segmenting an implant according to claim 1, wherein prior to the smoothing step (S5), a dynamic linearization step (S2) is applied to the image, for expanding the dynamic range of pale zones relative to the dynamic range of dark zones in the image.

11. A method of segmenting an implant according to claim 10, wherein the dynamic linearization step (S2) applies a cumulative distribution function to the histogram of the image.

12. A method of segmenting an implant according to claim 1, wherein prior to the smoothing step (S5), at least one anisotropic filtering step (S3) is applied to the image in order to decrease noise.

13. A method of segmenting an implant according to claim 12, wherein the anisotropic filtering step (S3) applies a filter to the preceding pixels u(s,t) of the image, which filter delivers at the following iteration for the pixel of coordinates s, a new pixel u(s,t+Δ

t) of coordinates s, defined by the equation;

u(s,t+Δ

t)=u(s,t)+Δ

t*div(f(∥

∇

u(s,t)∥

)·

∇

u(s,t))

where;

∇

designates the gradient operator;

div designates the divergence operator;

f is a decreasing non-uniformity function; and

Δ

t is a positive and non-zero predetermined increment.

14. A method of segmenting an implant according to claim 13, wherein the decreasing non-uniformity function f is a Gaussian function.

15. A method of segmenting an implant according to claim 14, wherein the new pixel u_s^(t+1) of the following iteration is given by:

16. A method of segmenting an implant according to claim 13, wherein the predetermined increment Δ

t is greater than 0 and less than 0.25.

17. A method of segmenting an implant according to claim 1, wherein a mask region is grown (S9) inside the boundaries of the resulting segmented portion, the region being grown from at least one selected starting point, by iterating at least one growth step, each growth step including within the mask region those pixels that are adjacent to the region obtained at the preceding iteration and that have brightness values that are sufficiently close to those of the region obtained at the preceding iteration.

18. A method of segmenting an implant according to claim 17, wherein the starting point for growth of the mask region is obtained (S10) by:

simple thresholding of the image above a brightness threshold (THR) that is less than the maximum brightness of the pixels of the image so as to obtain a first image zone; and

eroding and then dilatating the first zone so as to obtain a second zone, with the center of gravity of the second zone then being taken as the starting point for growing the mask region.

19. A method of segmenting an implant according to claim 1 and for correcting a digital mammogram of a breast containing an implant, information being represented in the mammogram by pixels of brightness greater than or less than a background, the method comprising performing the following steps in a computer:

detecting, in the histogram (H) of brightnesses (X) of the mammogram, an adipose first peak (F), calculating a first brightness value (θ

_F) corresponding to the adipose first peak (F), detecting a second peak (PI) in the histogram (H) of brightnesses (X) of the mammogram, said second peak (PI) corresponding to the presence of the implant, and calculating a second brightness value (θ

₂) corresponding to the implant second peak (PI);

resolving the mammogram into a low frequency image component (X_LP) and a high frequency image component (X_HP);

applying a first correction function (α

) to the low frequency image component (X_LP), in order to obtain a processed low frequency image component (X′

_LP), the first correction function (α

) comprising;

in the first configuration where breast information is represented in the mammogram by pixels of brightness greater than that of a background, an implant attenuation (ATT) for brightnesses situated at least at and around the implant peak (PI) relative to a first function (P1) applied to the other brightnesses situated between the first value (θ

_F) and the second value (θ

₂);

in the second configuration in which breast information is represented in the mammogram by pixels having brightnesses lower than that of a background, an implant amplification for brightnesses situated at least at and around the implant peak (PI) relative to a first function (P1) applied to the other brightnesses situated between the first value (θ

_F) and the second value (θ

₂); and

restoring the corrected mammogram on the basis of the high frequency component (X_HP) and of the processed low frequency component (X′

_LP).

20. A method of segmenting an implant according to claim 19, wherein:

in the first configuration, the attenuation (ATT) of the first correction function (α

) is applied to the brightnesses of the low frequency image component (X_LP) that are greater than or equal to a predetermined third brightness value (θ

₁) lying between the first brightness value (θ

_F) of the adipose peak (F) and the second brightness value (θ

₂) of the implant peak (PI); and

in the second configuration, the amplification of the first correction function (α

) is applied to the brightnesses of the low frequency image component (X_LP) that are lower than or equal to a predetermined third brightness value (θ

₁) lying between the first brightness value (θ

_F) of the adipose peak (F) and the second brightness value (θ

₂) of the implant peak (PI).

21. A method of segmenting an implant according to claim 20, wherein the third brightness value (θ

₁) is calculated as corresponding to the lowest valley detected in the brightness histogram (H) of the mammogram, for brightness values that are less than the second brightness value (θ

₂) of the implant peak (PI) in the first configuration and that are greater than the second brightness value (θ

₂) of the implant peak (PI) in the second configuration.

22. A method of segmenting an implant according to claim 20, wherein:

in the first configuration, the attenuation (ATT) of the first correction function (α

) for the low frequency component (X_LP) comprises a decreasing portion (ATT1) for brightnesses lying between the third brightness value (θ

₁) and the second brightness value (θ

₂) of the implant peak (PI), and a substantially constant portion (ATT2) for brightness values greater than or equal to the second brightness value (θ

₂) of the implant peak (PI); and

in the second configuration, the amplification of the first correction function (α

) of the low frequency component (X_LP) comprises a decreasing portion for brightnesses lying between the third brightness value (θ

₁) and the second brightness value (θ

₂) of the implant peak (PI), and a substantially constant portion for brightness values less than or equal to the second brightness value (θ

₂) of the implant peak (PI).

23. A method of segmenting an implant according to claim 22, wherein:

in the first configuration, the attenuation (ATT) of the first correction function (α

) for the low frequency component (X_LP) includes a substantially constant portion (ATT2) equal to the mean brightness value of the adipose peak (F) divided by the mean brightness value of the implant peak (PI); and

in the second configuration, the amplification of the first correction function (α

) for the low frequency component (X_LP) includes a substantially constant portion (ATT2) equal to the mean brightness value of the adipose peak (F) divided by the mean brightness value of the implant peak (PI).

24. A method of segmenting an implant according to claim 19, wherein said first correction function (α

) for the low frequency component (X_LP) includes a plateau (P1) multiplying brightness values lying between the first brightness value (θ

_F) of the adipose peak (F) and the third brightness value (θ

₁) by a coefficient that is substantially constant.

25. A method of segmenting an implant according to claim 19, wherein the first correction function (α

) for the low frequency image component (X_LP) includes;

in the first configuration, a raising (R) for brightness values that are less than the first value (θ

_F) of the adipose peak (F); and

in the second configuration, additional attenuation for brightness values greater than the first value (θ

_F) of the adipose peak (F).

26. A method of segmenting an implant according to claim 25, wherein:

in the first configuration, the raising (R) is a decreasing function for brightness values less than the first value (θ

_F) of the adipose peak (F); and

in the second configuration, the additional attenuation is a decreasing function for brightness values greater than the first value (θ

_F) of the adipose peak (F).

27. A method of segmenting an implant according to claim 19, wehrein a second correction function (β

) is applied to the high frequency image component (X_HP) in order to obtain a processed high frequency image component (X′

_HP), the second correction function (β

) comprising;

in the first configuration, an attenuation (AHP) for brightness values less than or equal to the first brightness value (θ

_F) of the adipose peak (F) relative to a second function applied to brightness values greater than said first brightness value (θ

_F); and

in the second configuration, an amplification for brightness values greater than or equal to the first brightness value (θ

_F) of the adipose peak (F) relative to a second function applied to brightness values that are less than said first brightness value (θ

_F).

28. Apparatus for segmenting an implant and for correcting a digital mammogram of a breast containing an implant, information being represented in the mammogram by pixels having brightnesses that are higher or lower than a background, the apparatus comprising a computer having:

means for detecting in the brightness (X) histogram (H) of the mammogram, an adipose first peak (F), means for calculating a first brightness value (θ

_F) corresponding to the adipose first peak (F), means for detecting in the brightness (X) histogram (H) of the mammogram, a second peak (PI) corresponding to the presence of the implant, and means for calculating a second brightness value (θ

₂) corresponding to the implant second peak (PI);

means for resolving the mammogram into a low frequency image component (X_LP) and into a high frequency image component (X_HP);

means for applying a first correction function (α

) to the low frequency image component (X_LP) in order to obtain a processed low frequency image component (X′

_LP), the first correction function (α

) comprising;

in the first configuration where breast information is represented in the mammogram by pixels of brightness greater than that of a background, an implant attenuation (ATT) for brightnesses situated at and around the implant peak (PI) relative to a first function (P1) applied to the other brightnesses situated between the first value (θ

_F) and the second value (θ

₂);

in the second configuration in which breast information is represented in the mammogram by pixels having brightnesses lower than a background, an implant amplification for brightnesses situated at least at and around the implant peak (PI) relative to a first function (P1) applied to the other brightnesses situated between the first value (θ

_F) and the second value (θ

₂); and

means for restoring the corrected mammogram on the basis of the high frequency component (X_HP) and of the processed low frequency component (X′

_LP).