Method and system for ground glass nodule (GGN) segmentation using a Markov random field and shape analysis

• US 7,289,653 B2
• Filed: 12/13/2004
• Issued: 10/30/2007
• Est. Priority Date: 07/31/2003
• Status: Expired due to Fees

• Current Assignee: Siemens Medical Solutions USA Incorporated (Siemens AG)
• Original Assignee: Siemens Corporate Research Incorporated (Siemens AG)
• Inventors: Fang, Ming, Zhang, Li
• Primary Examiner(s): Johns; Andrew W.
• Application Number:
  US11/010,813
• Publication Number:
  US 20050254697A1
• Time in Patent Office: 1,051 Days
• Field of Search: 382/128, 382/131, 382/132, 382/173, 128/922
• US Class Current: 382/131
• CPC Class Codes: G06T 2207/10081 Computed x-ray tomography [CT]
• CPC Class Codes: G06T 2207/20132 Image cropping
• CPC Class Codes: G06T 2207/30064 Lung nodule
• 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

Abstract

A system and method for ground glass nodule (GGN) segmentation is provided. The method comprises: selecting a point in a medical image, wherein the point is located in a GGN; defining a volume of interest (VOI) around the point, wherein the VOI comprises the GGN; removing a chest wall from the VOI; obtaining an initial state for a Markov random field; segmenting the VOI, wherein the VOI is segmented using the Markov random field; identifying vessels in the volume of interest; segmenting the vessels; and removing segmented vessels from the segmented ground glass nodule.

Claims

1. A method for ground glass nodule (GGN) segmentation, comprising:

selecting a point in a medical image, wherein said point is located in a GGN;

defining a volume of interest (VOI) around said point, wherein said VOI comprises said GGN;

obtaining an initial state for a Markov random field;

segmenting said VOI, wherein said VOI is segmented using said Markov random field;

identifying vessels in the volume of interest;

segmenting the vessels; and

removing segmented vessels from the segmented ground glass nodule.

2. A method for ground glass nodule (GGN) segmentation, comprising:

selecting a point in a medical image, wherein said point is located in a GGN;

defining a volume of interest (VOI) around said point, wherein said VOI comprises said GGN;

removing a chest wall from said VOI;

obtaining an initial state for a Markov random field;

segmenting said VOI, wherein said VOI is segmented using said Markov random field; and

performing a shape analysis to remove blood vessels attached to or near said GGN after said VOI has been segmented using said MarKov random field, wherein said step of performing a shape analysis comprises;

identifying vessels in said VOI;

segmenting said vessels; and

removing segmented vessels from said segmented GGN.

3. A method as recited in claim 2, wherein said step of identifying said vessels comprises performing a thresholding and connected component analysis.

4. A method as recited in claim 2, wherein said step of identifying said vessels comprises performing a 3D compactness analysis.

5. A method as recited in claim 2, wherein said step of segmenting said vessels comprises performing intensity constrained dilation.

6. A method as recited in claim 2, including a step of performing a morphological operation following said step of removing segmented vessels.

7. A method as recited in claim 2, including a step of removing any chest wall from said VOI.

8. A method as recited in claim 7, wherein said step of removing a chest wall from said VOI comprises distinguishing said chest wall from vessels based on the fact that said chest wall is always located on a corner of said VOI.

9. A method as recited in claim 7, wherein said step of removing a chest wall from said VOI comprises utilizing a connected component technique to remove an area that belongs to said chest wall.

10. A method as recited in claim 2, comprising a step of morphological opening to remove small branches, which are smaller than said GGN, resulting from said step of removing segmented vessels.

11. A method as recited in claim 2, including a step of detecting said GGN utilizing a computer-aided GGN detection technique.

12. A method as recited in claim 2, including a step of detecting said GGN utilizing a manual technique for detecting said GGN.

13. A method as recited in claim 2, wherein said GGN comprises a pure GGN and a mixed GGN.

14. A method as recited in claim 2, wherein said step of identifying vessels comprises:

applying thresholding with a sufficiently high threshold T_high, so as to identify vessels and thereby derive a thresbolded VOI;

utilizing a process of connected component labeling on said thresholded VOI to find high intensity regions;

utilizing 3-dimensional (3D) compactness of said high intensity regions to distinguish vessels from solid components in said GGN; and

applying thresholding with a low threshold T_low, lower than said high threshold T_high, to obtain a more complete vessel segmentation.

15. A method as recited in claim 14, wherein said step of segmenting vessels VOI comprises applying an intensity constrained dilation so as to include boundary voxels into said vessel segmentation.

16. A method as recited in claim 14, wherein said step of segmenting vessels comprises applying an intensity constrained dilation so as to include boundary voxels into said vessel segmentation and obtain an essentially complete vessel segmentation O_v as follows:

O_v=O_vc∪

{ x| x∈

(O_vc⊕

SE−

O_vc) and I_x≧

T_low},

wherein O_vc is the vessel core part segmentation obtained from the foregoing step wherein vessels are distinguished from solid components in said GGN, ⊕

denotes morphological dilation, and SE is a dilation structure element, I_x is an intensity value at location x, and T_low is said lower threshold; and

said step of removing segmented vessels from said initially segmented GGN comprises removing said essentially complete vessel segmentation O_v from said segmented GGN obtained in said step of segmenting said VOI using a Markov random field (MRF) procedure, so as to obtain a final GGN segmentation without vessels O_onlyggo=O_inilggo−

O_v.

17. A method as recited in claim 15, wherein said dilation structure element is 3×

3×

3 voxel cube.

18. A method as recited in claim 2 comprising:

applying a morphological opening operation, constrained by intensity, to O_onlyggn so as to obtain a cleaned final GGN segmentation O_ggn, free of any remaining extra branches due to erroneous vessel inclusion resulting from a GGN intensity model including lower intensities than boundary vessel voxels, as follows;

O_ggn=(O_onlyggn∘

SE)−

({x| x∈

(O_onlyggn∘

SE)}∩

{x| x∉

O_onlyggn}∩

{ x|I_x<

T_low}),

wherein ∘

denotes morphological opening, SE is a structuring element, and wherein in the foregoing equation, the part before the minus sign is the general form of morphological opening, and the part following the minus sign imposes an intensity constraint such that if a pixel added by dilation is lower than the threshold T_low, then this pixel is removed from said cleaned final segmentation.

19. A method as recited in claim 18, wherein said structuring element SE is a 2D disk.

20. A method as recited in claim 2, comprising:

acquiring said medical image.

21. A method as recited in claim 20, wherein said medical image is acquired using a computed tomographic (CT) imaging technique.

22. A method as recited in claim 2, wherein said point is automatically selected.

23. A method as recited in claim 2, wherein said point is manually selected.

24. A method as recited in claim 2, further comprising:

detecting said GGN using a computer-aided GGN detection technique.

25. A method as recited in claim 2, further comprising:

detecting said GGN manually.

26. A method as recited in claim 2, wherein said GGN is one of a pure GGN and a mixed GGN.

27. A method as recited in claim 2, wherein said chest wall is removed by performing a connected component analysis.

28. A method as recited in claim 2, further comprising:

defining one of a shape and a size of said VOI.

29. A method as recited in claim 2, wherein the initial state for said Markov random field is obtained by performing a region growing on said VOI after said chest wall is removed.

30. A method as recited in claim 2, wherein the step of segmenting said VOI using said Markov random field comprises:

defining a posteriori probability for said VOI; and

labeling each pixel in said VOI to obtain a maximum of the posteriori probability, wherein each pixel in said VOI is labeled as one of a GGN and a background.

31. A method as recited in claim 30, wherein said defined posteriori probability is computed by

P(L|F)∝

P(F|L)P(L).

32. A method as recited in claim 30, wherein the step of labeling each pixel is computed by

$l_{\stackrel{\_}{x}}$

( i ) = arg ⁢

⁢

min l ∈

{ g , b } ⁢

{ U ⁡

( g , i - 1 ) + 1 2 ⁢

σ

2 ⁡

[ f ⁡

( x _ ) - μ

g ] , U ⁡

( b , i - 1 ) + 1 2 ⁢

σ

2 ⁡

[ f ⁡

( x _ ) - μ

b ] } .

33. A method as recited in claim 30, wherein said labeling comprises:

scanning said VOI until a convergence is reached.

34. A method as recited in claim 2, fUrther comprising:

displaying said VOI segmented using said Markov random field.

35. A system for ground glass nodule (GGN) segmentation, comprising:

a memory device for storing a program; and

a processor in communication with said memory device, said processor operative with said program to;

define a volume of interest (VOI) around a GGN using data associated with a medical image of a lung,

obtain an initial state for a Markov random field,

segment said VOI, wherein said VOI is segmented using said Markov random field, and

perform a shape analysis to remove blood vessels attached to said GGN in said VOI segmented using said Markov random field.

36. A system as recited in claim 35, wherein said processor being operative to perform said shape analysis includes said processor being operative with said program to:

distinguish vessels from said VOI,

segment vessels in said initially segmented VOI, and

remove segmented vessels from said segmented GGN in said VOI.

37. A system as recited in claim 36, wherein said processor is operative with said program to remove a chest wall from said VOI.

38. A system as recited in claim 37, wherein said processor is operative with said program to distinguish said chest wall from vessels.

39. A system as recited in claim 38, wherein said processor is operative with said program to distinguish said chest wall from vessels based on the fact that said chest wall is always located on a corner of said VOI.

40. A system as recited in claim 37, wherein said processor is operative with said program to utilize a connected component technique to remove a region that belongs to said chest wall.

41. A system as recited in claim 37, wherein said processor is further operative with said program code to remove said chest wall by performing a connected component analysis.

42. A system as recited in claim 36, wherein said processor is operative with said program to apply a morphological opening operation to remove small branches, less than the size of said GGN, resulting from said step of removing vessels from said segmented GGN in said VOI.

43. A system as recited in claim 42, wherein said processor is operative with said program to detect said GGN utilizing a computer-aided GGN detection technique.

44. A system method as recited in claim 42, wherein said processor is operative with said program to detect said GGN utilizing a manual technique for detecting said GGN.

45. A system as recited in claim 36, wherein said GGN comprises a pure GGN and a mixed GGN.

46. A system as recited in claim 36, wherein said processor being operative with said program:

applies thresholding with a sufficiently high threshold T_high, so as to identify vessels and thereby derive a thresholded VOI;

utilizes a process of connected component labeling on said thresholded VOI to find high intensity regions;

utilizes a process of compactness analysis to identify vessels from high intensity regions;

utilizes 3-dimensional (3D) compactness of said high intensity regions to distinguish vessels from solid components in said GGN; and

applies thresholding with a low threshold T_low, lower than said high threshold T_high, to obtain a more complete vessel segmentation.

47. A system as recited in claim 46, wherein said processor, being operative with said program applies an intensity constrained dilation so as to include boundary voxels into said vessel segmentation.

48. A system as recited in claim 46, wherein said processor, being operative with said program, applies an intensity constrained dilation so as to include boundary voxels into said vessel segmentation and obtain an essentially complete vessel segmentation O_v as follows:

O_v=O_vc∪

{ x| x∈

(O_vc⊕

SE−

O_vc) and I_x≧

T_low},

wherein O_vc is the vessel core part segmentation obtained from the foregoing operation wherein vessels are distinguished from solid cdmponents in said GGN, ⊕

denotes morphological dilation, and SE is a dilation structure element, I_x is an intensity value at location x, and T_low is said lower threshold; and

said step of removing segmented vessels from said initially segmented GGN comprises removing said essentially complete vessel segmentation O_v from an initial GGN segmentation O_initggo of said GGN, so as to obtain GGN segmentation without vessels O_onlyggo=O_initggo−

O_v.

49. A system as recited in claim 48, wherein said dilation structure element is 3×

3×

3 pixel cube.

50. A system as recited in claim 46, wherein said processor is operative with said program to apply a morphological opening operation, constrained by intensity, to O_onlyggn so as to obtain a cleaned final GGN segmentation O_ggn, free of any remaining extra branches due to erroneous vessel inclusion resulting from a GGN intensity model including lower intensities than boundary vessel voxels, as follows:

O_ggn=(O_onlyggn∘

SE)−

({x| x∈

(O_onlyggn∘

SE)}∩

{x| x∉

O_onlyggn}∩

{ x|I_x<

T_low}),

wherein ∘

denotes morphological opening, SE is a structuring element, and wherein in the foregoing equation, the part before the minus sign is the general form of morphological opening, and the part following the minus sign imposes an intensity constraint such that, if a pixel added by dilation is lower than the threshold T_low, then this pixel is removed from said cleaned final segmentation.

51. A system as recited in claim 50, wberein said processor is further operative with said program code to utilize a 2D disk as said structuring element SE.

52. A system as recited in claim 36, wherein said processor is further operative with said program code to:

acquire said medical image, wherein said medical image is acquired using a computed comographic (CT) imaging technique.

53. A system as recited in claim 36, wherein said processor is further operative with said program code to obtain the initial state for the Markov random field by performing a region growing on said VOI after said chest wall is removed.

54. A system as recited in claim 36, wherein said processor is further operative with said program code when segmenting said VOI using said Markov random field to:

define aposteriori probability for said VOI; and

label each pixel in said VOI to obtain a maximum of the posteriori probability, wherein each pixel in said VOI is labeled as one of a GGN and a background.

55. A system as recited in claim 54, wherein said processor is further operative with said program code to:

compute said defined posteriori probability by

P(L|F)∝

P(F|L)P(L).

56. A system as recited in claim 55, wherein said processor is further operative with said program code to compute said step of labeling each pixel by:

$l_{\stackrel{\_}{x}}<br><br>\left(i\right)=\underset{l\in <br><br>\left\{g,b\right\}}{\arg <br><br><br><br>\min }<br><br>\left\{\begin{array}{c}U<br><br>\left(g,i-1\right)+\frac{1}{2<br><br>{\mathrm{\sigma <br><br>}}^2}<br><br>\left[f<br><br>\left(\stackrel{\_}{x}\right)-{\mathrm{\mu <br><br>}}_g\right],\\ U<br><br>\left(b,i-1\right)+\frac{1}{2<br><br>{\mathrm{\sigma <br><br>}}^2}<br><br>\left[f<br><br>\left(\stackrel{\_}{x}\right)-{\mathrm{\mu <br><br>}}_b\right]\end{array}\right\}.$

57. A system as recited in claim 36, wherein said processor is further operative with said program code to:

display said VOI segmented using said Markov random field, wherein said GGN is visible.

58. A computer program product comprising a computer useable medium having computer program logic recorded thereon for ground glass nodule (GGN) segmentation, said computer program logic comprising:

program code for selecting a point in a medical image, wherein said point is located in or near a GGN;

program code for defining a volume of interest (VOI) around said point, wherein said VOI comprises said GGN;

program code for removing a chest wall from said VOI;

program code for obtaining an initial state for a Markov random field; and

program code for segmenting said VOI, wherein said VOI is segmented using said Markov random field;

identifying vessels from said VOI,

segmenting vessels in said VOI, and

removing segmented vessels from said VOI.

59. A system for ground glass nodule (GGN) segmentation, comprising:

means for selecting a point in a medical image, wherein said point is located in a GGN;

means for defining a volume of interest (VOI) around said point, wherein said VOI comprises said GGN;

means for removing a chest wall from said VOI;

means for obtaining an initial state for a Markov random field;

means for segmenting said VOI, wherein said VOI is segmented using said Markov random field;

means for identifying vessels from said VOI;

means for segmenting vessels in said VOI; and

means for removing segmented vessels from said VOI.

60. A method for ground glass nodule (GGN) segmentation in pulmonary computed tomographic (CT) volumes using a Markov random field, comprising:

selecting a GGN from data associated with a pulmonary CT volume;

defining a volume of interest (VOI) around said GGN;

removing a chest wall from said VOI by performing a connected component analysis on said VOI;

obtaining an initial state for an iterated condition mode (ICM) procedure by segmenting said VOI after said chest wall is removed; and

segmenting said VOI using a Markov random field, wherein the segmentation comprises;

defining aposteriori probability for said VOI,

performing said ICM procedure, wherein said ICM procedure comprises labeling each pixel in said VOI to obtain maximum of the posteriori probability, wherein each pixel in said VOI is labeled as one of a GGN and a background until the label of each pixel in said VOI is not changed any more,

identifying vessels from said VOI,

segmenting vessels in said VOI, and

removing segmented vessels from said VOI.

61. A method as recited in claim 60, wherein the defined posteriori probability is computed by

P(L|F)∝

P(F|L)P(L).

62. A method as recited in claim 60, wherein said step of labeling each pixel during said ICM procedure is computed by

$l_{\stackrel{\_}{x}}$

( i ) = arg ⁢

⁢

min l ∈

{ g , b } ⁢

{ U ⁡

( g , i - 1 ) + 1 2 ⁢

σ

2 ⁡

[ f ⁡

( x _ ) - μ

g ] , U ⁡

( b , i - 1 ) + 1 2 ⁢

σ

2 ⁡

[ f ⁡

( x _ ) - μ

b ] } .

63. A method as recited in claim 60, wherein said ICM procedure begins from said initial state.

64. A method as recited in claim 63, comprising:

applying a morphological opening operation to remove small branches resulting from said step of removing segmented vessels from said initially segmented GGN;

applying thresholding with a sufficiently high threshold T_high, so as to identify vessels and thereby derive a thresholded VOI;

utilizing a process of connected component labeling on the thresbolded VOI to find high intensity regions;

utilizing 3-dimensional (3D) compactness of said high intensity regions to distinguish vessels from solid components in said GGN; and

applying intensity constrained dialation with a low threshold T_low, lower than said high threshold T_high, to obtain a more complete vessel segmentation.

65. A method as recited in claim 64, wherein said step of segmenting vessels in said VOI comprises applying an intensity constrained dilation so as to include boundary voxels into said vessel segmentation.

66. A method as recited in claim 64, wherein said step of segmenting vessels in said VOI comprises applying an intensity constrained dilation so as to include boundary voxels into said vessel segmentation and obtain an essentially complete vessel segmentation O_v as follows:

O_v=O_vc∪

{ x| x∈

(O_vc⊕

SE−

O_vc) and I_x≧

T_low},

wherein O_vc is the vessel core part segmentation obtained from the foregoing step wherein vessels are distinguished from solid components in said GGN, ⊕

denotes morphological dilation, and SE is a dilation structure element, I_x is an intensity value at location x, and T_low said lower threshold; and

said step of removing segmented vessels from said initially segmented GGN comprises removing said essentially complete vessel segmentation O_v from an initial GGN segmentation O_initggo of said an initially segmented GGN, so as to obtain GGN segmentation without vessels O_onlyggo=O_initggo−

O_v.

67. A method as recited in claim 66, including a step of removing a chest wail from said VOI.

69. A method as recited in claim 67, wherein said step of removing a chest wall from said VOI comprises utilizing a connected component analysis technique to remove a region that belongs to said chest wall.

70. A method as recited in claim 67, including a step of detecting said GGN utilizing a computer-aided GGN detection technique.

71. A method as recited in claim 64, including a step of detecting said GGN utilizing a manual technique for detecting said GGN.

72. A method as recited in claim 64, wherein said GGN comprises a pure GGN and a mixed GGN.

73. A method as recited in claim 64, wherein said step of applying a morphological opening operation comprises:

applying morphological opening operation, constrained by intensity, to O_onlyggn so as to obtain a cleaned final GGN segmentation O_ggn, free of any remaining extra branches due to erroneous vessel inclusion resulting from a GGN intensity model including lower intensities than boundary vessel voxels, as follows;

O_ggn=(O_onlyggn∘

SE)−

({x| x∈

(O_onlyggn∘

SE)}∩

{x| x∉

O_onlyggn}∩

{ x|I_x<

T_low}),

wherein ∘

denotes morphological opening, SE is a structuring element, and wherein in the foregoing equation, the part before the minus sign is the general form of morphological opening, and the part following the minus sign imposes an intensity constraint such that if a pixel added by dilation is lower than the threshold T_low, then this pixel is removed from said cleaned final segmentation

74. A method as recited in claim 73, wherein said structuring element SE is a 2D disk.

75. A method as recited in claim 74, including a step of removing a chest wall from said VOI.

76. A method as recited in claim 60, wherein said step of defining a volume of interest comprises defining the shape and size of said VOI.

77. A method as recited in claim 60, wherein said step of segmenting vessels comprises:

applying thresholding with a sufficiently high threshold T_high, so as to identify vessels and thereby derive a thresholded VOI;

utilizing a process of connected component labeling on said thresholded VOI to find high intensity regions;

utilizing 3-dimensional (3D) compactness of said high intensity regions to distinguish vessels from solid components in said GGN; and

applying intensity constrained dilation with a low threshold T_low, lower than said high threshold T_high, to obtain a more complete vessel segmentation.

78. A method as recited in claim 60, wherein said step of segmenting vessels comprises applying an intensity constrained dilation so as to include boundary voxels into said vessel segmentation.

79. A method as recited in claim 60, wherein said step of segmenting vessels comprises applying an intensity constrained dilation so as to include boundary voxels into said vessel segmentation and obtain an essentially complete vessel segmentation O_v as follows:

O_v=O_vc∪

{ x| x∈

(O_vc⊕

SE−

O_vc) and I_x≧

T_low},

wherein is the vessel core part segmentation obtained from the foregoing step wherein vessels are distinguished from solid components in said GGN, ⊕

denotes morphological dilation, and SE is a dilation structure element, I_x is an intensity value at location x, and T_low said lower threshold; and

said step of removing segmented vessels from said initially segmented VOI comprises removing said essentially complete vessel segmentation O_v from an initial GGN segmentation O_initggo of said VOI, so as to obtain GGN segmentation without vessels O_onlyggo=O_initggo−

O_v.

80. A method as recited in claim 79, wherein said dilation structure element is 3×

3×

3 voxel cube.

81. A method as recited in claim 60, comprising:

applying a morphological opening operation, constrained by intensity, to O_onlyggn so as to obtain a cleaned final GGN segmentation O_ggn, free of any remaining extra branches due to erroneous vessel inclusion resulting from a GGN intensity model including lower intensities than boundary vessel voxels, as follows;

O_ggn=(O_onlyggn∘

SE)−

({x| x∈

(O_onlyggn∘

SE)}∩

{x| x∉

O_onlyggn}∩

{ x|I_x<

T_low}),

wherein ∘

denotes morphological opening, SE is a structuring element, and wherein in the foregoing equation, the part before the minus sign is the general form of morphological opening, and the part following the minus sign imposes an intensity constraint such that if a pixel added by dilation is lower than the threshold T_low, then this pixel is removed from said cleaned final segmentation.

82. A method as recited in claim 81, wherein said structuring element SE is a 2D disk.

68. A method as recited in caim 67, wherein said step of removing a chest wall from said VOI comprises distinguishing said chest wall from vessels based on the fact that said chest wall is always located on a corner of said VOI.