System and method for detecting tagged material using alpha matting

• US 7,876,947 B2
• Filed: 10/02/2008
• Issued: 01/25/2011
• Est. Priority Date: 10/10/2007
• Status: Active Grant

• Current Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Original Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Inventors: Lakare, Sarang, Lee, Tin Man, Wolf, Matthias
• Primary Examiner(s): Tabatabai; Abolfazl
• Application Number:
  US12/244,428
• Publication Number:
  US 20090097728A1
• Time in Patent Office: 845 Days
• Field of Search: 382/128, 382/129, 382/130, 382/131, 382/132, 382/133, 382/134, 382/171, 382/173, 382/190, 378/4, 378/8, 378 21- 27, 378/98.4, 378/98.6, 378/101, 378/901, 600/407, 600/410, 600/411, 600/425, 128/915, 128/920, 128/922
• US Class Current: 382/131
• CPC Class Codes: G06T 2207/10081 Computed x-ray tomography [CT]
• CPC Class Codes: G06T 2207/30032 Colon polyp
• CPC Class Codes: G06T 2207/30096 Tumor; Lesion
• 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: G06T 7/155 involving morphological operators

Abstract

A method for computer-aided object classification, soft segmentation and layer extraction in computed tomographic colonography includes providing a contrast enhanced computed tomography (CT) digital image of the colon, finding a foreground region of voxels with an intensity higher than a pre-defined foreground threshold, creating a 3D trimap of the colon where the image is segmented into the foreground region, a background region, and an unknown region between the foreground and background, starting from the background, extracting successive layers of the unknown region until the foreground region is reached, and classifying each extracted layer as background or foreground, and generating a foreground matte, a background matte, and an alpha matte, where alpha indicates a mixing ration of foreground and background voxels.

Claims

1. A method for computer-aided object classification, soft segmentation and layer extraction in computed tomographic colonography, said method comprising the steps of:

providing a contrast enhanced computed tomography (CT) digital image of the colon, said image comprising a plurality of intensities associated with a 3-dimensional (3D) grid of voxels;

finding a foreground region of voxels with an intensity higher than a pre-defined foreground threshold;

creating a 3D trimap of said colon wherein said image is segmented into said foreground region, a background region, and an unknown region between the foreground and background;

starting from said background, extracting successive layers of said unknown region until said foreground region is reached, and classifying each extracted layer as background or foreground; and

generating a foreground matte, a background matte, and an alpha matte, wherein alpha indicates a mixing ration of foreground and background voxels.

2. The method of claim 1, wherein said background region voxels represent either regions of air in said colon, or regions of known soft tissue in said colon.

3. The method of claim 2, wherein creating a 3D trimap of said colon comprises morphologically dilating the 3D foreground region one voxel at a time until background region voxels are reached, wherein a region through which said foreground was diluted is a region of unknown type, and categorizing said background region voxels as either air or soft tissue.

4. The method of claim 1, wherein classifying each extracted layer as background or foreground comprises forming a 3D sphere around a candidate layer voxel and neighboring known background voxels, and analyzing closest known background voxels to determine whether said candidate layer voxel is a background air voxel or a background soft tissue voxel.

5. The method of claim 4, further comprising calculating said mixing ratio alpha for each voxel in said unknown region from alpha=(C−

B)/(F−

B), wherein C is the intensity of said voxel, B is a background intensity, and F is a foreground intensity.

6. The method of claim 5, wherein analyzing closest known background voxels comprises finding values of F and B that maximize a sum of log-likelihoods L(C|B, F, α

)+L(F)+L(B)+L(α

), wherein

7. The method of claim 1, wherein generating a foreground matte, a background matte, and an alpha matte, comprises:

forming a 3D sphere about each candidate voxel in said region of unknown type;

obtaining from the background region within the sphere a number of air voxels, a number of tissue voxels, their respective distance from the candidate voxel, and their variance;

obtaining from the foreground region within the sphere a number of tagged voxels having an intensity above a threshold for contrast enhancement, their respective distance from the candidate, and their variance;

weighting the intensity of each tissue voxel by its distance from said candidate voxel;

weighting the intensity of each air voxel by its distance from said candidate voxel;

weighting the intensity of each foreground voxel by its distance from said candidate voxel; and

calculating alpha from

C=α

F+∫

<sub>Ω

(air)</sub>H_air(1−

α

(i))B_air(i)+∫

<sub>Ω

(tissue)</sub>H_tissue(1−

α

(i))B_tissue(i), wherein C is the intensity of said candidate voxel, B_air is a background air intensity, B_tissue is a background tissue intensity, F is a foreground intensity, Ω

is a region of integration in either air voxels or tissue voxels, i represents a voxel in the region of integration, and H is the Heaviside function having a value of 1 inside the region of integration.

8. The method of claim 7, wherein a foreground weighting function is:

9. The method of claim 7, wherein a candidate voxel is identified as a T-junction where the number of tissue voxels, air voxels, and a number of tagged voxels are greater than zero inside a sphere of a predetermined small radius.

10. The method of claim 7, wherein for tagged voxels laying on a haustral fold, the method further comprises:

dilating said tagged voxels by a number of voxels less than an original size of said morphological dilation;

removing the tagged voxels and dilated voxels from a voxel window whose side is greater than the original size of said morphological dilation;

performing region growing to find connected voxels;

identifying the connected voxels; and

performing matting, wherein there is no type change of any voxel that has been identified as a connected voxel.

11. A method for computer-aided object classification, soft segmentation and layer extraction in computed tomographic colonography, said method comprising the steps of:

providing a contrast enhanced computed tomography (CT) digital image of the colon, said image comprising a plurality of intensities associated with a 3-dimensional (3D) grid of voxels;

creating a 3D trimap of said colon wherein said image is segmented into a foreground region, a background region, and an unknown region between the foreground and background;

forming a 3D sphere about a candidate voxel in said region of unknown type;

obtaining from the background region within the sphere a number of air voxels, a number of tissue voxels, their respective distance from the candidate voxel, and their variance;

obtaining from the foreground region within the sphere a number of tagged voxels having an intensity above a threshold for contrast enhancement, their respective distance from the candidate, and their variance;

weighting the intensity of each tissue voxel by its distance from said candidate voxel;

weighting the intensity of each air voxel by its distance from said candidate voxel;

weighting the intensity of each foreground voxel by its distance from said candidate voxel; and

calculating a mixing ratio of foreground and background voxels a from

C=α

F+∫

<sub>Ω

(air)</sub>H_air(1−

α

(i))B_air(i)+∫

<sub>Ω

(tissue)</sub>H_tissue(1−

α

(i))B_tissue(i), wherein C is the intensity of said candidate voxel, B_air is a background air intensity, B_tissue is a background tissue intensity, F is a foreground intensity, Ω

is a region of integration in either air voxels or tissue voxels, i represents a voxel in the region of integration, and H is the Heaviside function having a value of 1 inside the region of integration.

12. The method of claim 11, wherein said foreground region comprises voxels with an intensity higher than a pre-defined foreground threshold for contrast enhanced voxels.

13. The method of claim 11, further comprising, starting from said background, extracting successive layers of said unknown region until said foreground region is reached, and classifying each extracted layer as background or foreground.

14. The computer readable program storage device of claim 13, wherein said background region voxels represent either regions of air in said colon, or regions of known soft tissue in said colon.

15. The computer readable program storage device of claim 14, wherein creating a 3D trimap of said colon comprises morphologically dilating the 3 D foreground region one voxel at a time until background region voxels are reached, wherein a region through which said foreground was diluted is a region of unknown type, and categorizing said background region voxels as either air or soft tissue.

20. The computer readable program storage device of claim 14, wherein generating a foreground matte, a background matte, and an alpha matte, comprises:

forming a 3D sphere about each candidate voxel in said region of unknown type;

obtaining from the background region within the sphere a number of air voxels, a number of tissue voxels, their respective distance from the candidate voxel, and their variance;

obtaining from the foreground region within the sphere a number of tagged voxels having an intensity above a threshold for contrast enhancement, their respective distance from the candidate, and their variance;

weighting the intensity of each tissue voxel by its distance from said candidate voxel;

weighting the intensity of each air voxel by its distance from said candidate voxel;

weighting the intensity of each foreground voxel by its distance from said candidate voxel; and

calculating alpha from

C=α

F+∫

<sub>Ω

(air)</sub>H_air(1−

α

(i) B_air(i)+∫

<sub>Ω

(tissue)</sub>H_tissue(1−

α

(i))B_tissue(i), wherein C is the intensity of said candidate voxel, B_air is a background air intensity, B_tissue is a background tissue intensity F is a foreground intensity, Ω

is a region of integration in either air voxels or tissue voxels, i represents a voxel in the region of integration, and H is the Heaviside function having a value of 1 inside the region of integration.

21. The computer readable program storage device of claim 20, wherein a foreground weighting function is:

22. The computer readable program storage device of claim 20, wherein a candidate voxel is identified as a T-junction where the number of tissue voxels, air voxels, and a number of tagged voxels are greater than zero inside a sphere of a predetermined small radius.

23. The computer readable program storage device of claim 20, wherein for tagged voxels laying on a haustral fold, the method further comprises:

dilating said tagged voxels by a number of voxels less than an original size of said morphological dilation;

removing the tagged voxels and dilated voxels from a voxel window whose side is greater than the original size of said morphological dilation;

performing region growing to find connected voxels;

identifying the connected voxels; and

performing matting, wherein there is no type change of any voxel that has been identified as a connected voxel.

16. A program storage device readable by a computer, tangibly embodying a program of instructions executable by the computer to perform the method steps for computer-aided object classification, soft segmentation and layer extraction in computed tomographic colonography, said method comprising the steps of:

providing a contrast enhanced computed tomography ((CT) digital image of the colon, said image comprising a plurality of intensities associated with a 3-dimensional (3D) grid of voxels;

finding a foreground region of voxels with an intensity higher than a pre-defined foreground threshold;

creating a 3D trimap of said colon wherein said image is segmented into said foreground region, a background region, and an unknown region between the foreground and background;

starting from said background, extracting successive layers of said unknown region until said foreground region is reached, and classifying each extracted layer as background or foreground; and

generating a foreground matte, a background matte, and an alpha matte, wherein alpha indicates a mixing ration of foreground and background voxels.

17. The computer readable program storage device of claim 16, wherein classifying each extracted layer as background or foreground comprises forming a 3D sphere around a candidate layer voxel and neighboring known background voxels, and analyzing closest known background voxels to determine whether said candidate layer voxel is a background air voxel or a background soft tissue voxel.

18. The computer readable program storage device of claim 17, the method further comprising calculating said mixing ratio alpha for each voxel in said unknown region from alpha=(C−

B)/(F−

B) wherein C is the intensity of said voxel, B is a background intensity, and F is a foreground intensity.

19. The computer readable program storage device of claim 18, wherein analyzing closest known background voxels comprises finding values of F and B that maximize a sum of log-likelihoods L(C|B, F, α

)+L(F)+L(B)+L(α

), wherein