Method for segmenting objects in images

• US 20080292169A1
• Filed: 05/21/2008
• Published: 11/27/2008
• Est. Priority Date: 05/21/2007
• Status: Active Grant

• Current Assignee: Cornell University
• Original Assignee: Cornell University
• Inventors: Wang, Yi, Lee, Hae-Yeou, Codella, Noel C. F.
• Granted Patent: US 8,369,590 B2
• Time in Patent Office: Days
• US Class Current: 382/131
• CPC Class Codes: G06T 2207/10072 Tomographic images
• CPC Class Codes: G06T 2207/10088 Magnetic resonance imaging [MRI]
• CPC Class Codes: G06T 2207/30048 Heart; Cardiac
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/174 involving the use of two or more images
• CPC Class Codes: G06T 7/62 of area, perimeter, diameter or volume

Abstract

A method for identifying an attribute of an object represented in an image comprising data defining a predetermined spatial granulation for resolving the object, where the object is in contact with another object. In an embodiment, the method comprises identifying data whose values indicate they correspond to locations completely within the object, determining a contribution to the attribute provided by the data, and identifying additional data whose values indicate they are not completely within the object. The method next interpolates second contributions to the attribute from the values of the additional data and finds the attribute of the object from the first contribution and second contributions. The attribute may be, for example, a volume, and the values may correspond, for example, to intensity.

Claims

1. A method for identifying an attribute of an object represented in an image comprising data defining a predetermined spatial granulation for resolving the object, where the object is in contact with another object, the method comprising:

identifying the data whose values indicate they correspond to locations completely within the object;

determining a first contribution to the attribute provided by the data;

identifying additional data whose values indicate they correspond to locations that are partially within the object;

interpolating second contributions to the attribute from the values of the additional data; and

finding the attribute of the object from the first contribution and second contributions.

2. The method of claim 1, wherein the data of the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

3. The method of claim 1a, further comprising correcting the image.

4. The method of claim 1, wherein the attribute is volume.

5. The method of claim 1, wherein the values correspond to intensity.

6. The method of claim 1, wherein identifying the data includes growing a region by identifying neighboring data whose values are within a predetermined amount of the value of a datum determined to have a value indicating that the datum corresponds to a location completely within the object.

7. The method of claim 1, wherein interpolating second contributions to the attribute from the values of the additional data includes interpolating numbers from a first characteristic of the object and a second characteristic of said another object.

8. The method of claim 7, wherein the first characteristic is a mean value of data whose values indicate they correspond to locations completely within the object and wherein the second characteristic is a mean value of data whose values indicate they correspond to locations completely within said another object.

9. A method for identifying an object in an image, the object in contact with another object, the method comprising:

identifying one or more partial voxels located at an edge of the object and encompassing a portion of the object and a portion of said another object; and

estimating an amount of the object encompassed by the one or more partial voxels using a first value of an attribute the one or more partial voxels.

10. The method of claim 9, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

11. The method of claim 9, further comprising correcting the image.

12. The method of claim 9, wherein the object has a second value of the attribute and said another object has a third value of the attribute and wherein estimating an amount of the object encompassed by the one or more partial voxels includes, interpolating the amount based on the first value, the second value, and the third value.

13. The method of claim 12, wherein the attribute corresponds to intensity.

14. A method for identifying a volume of an object in an image comprising voxels, the image including another object touching the object, the method comprising:

growing a region of one or more voxels in the image completely within the object;

continuing to grow the region by identifying one or more additional voxels, each additional voxel having an attribute whose value is different than a characteristic value of the attribute for either the object or said another object and having a neighborhood of adjacent voxels determined to be within the region.

15. The method of claim 14, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

16. The method of claim 14, further comprising correcting.

17. A method of claim 14, further comprising summing volumes of the voxels determined to be in the region in order to provide a value of a total volume of the object.

18. The method of claim 14, wherein the attribute corresponds to intensity.

19. A method for identifying an attribute of an object represented in an image comprising data, each datum corresponding to a region of the object as imaged and thereby defining a predetermined spatial granulation for resolving the image of the object, where the object is in contact with another object in the image data, the method comprising:

calculating a first characteristic of the data contained completely within the object;

calculating a second characteristic of data contained completely within said another object;

interpolating third characteristics of data corresponding to an area of the image generated from the data that overlaps both the object and said another object; and

using the third characteristics to determine contributions of the data overlapping the object and said another object in order to resolve a value of the attribute of the object beyond a resolution provided by the spatial granulation of the data.

20. The method of claim 19, wherein the object comprises a tissue type and said another object comprises another tissue type.

21. The method of claim 19, wherein the data is collected from one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

22. The method of claim 19, further comprising correcting the image.

23. The method of claim 19, wherein the first, second, and third characteristics are based on intensity values.

24. The method of claim 19, wherein the image comprises a plurality of slices and wherein the method further comprises selecting a seed in a slice and propagating the seed to remaining slices.

25. A method for measuring the volume of an object in an image, the object meeting another object, the method comprising:

growing a test object starting at a seed voxel within the object;

computing the volume of the test object based on a set of voxels having values of a common attribute close to a value of the seed voxel; and

stopping the growth of the test object when a rate of expansion exceeds a predetermined value.

26. The method of claim 25, wherein the object corresponds to a left ventricle of a heart organ and said another object corresponds to a myocardium containing the left ventricle.

27. The method of claim 25, wherein growing a test object is performed according to an iteratively changed threshold and wherein the rate of expansion is based on an approximation of a derivative of the volume.

28. The method of claim 27, further comprising computing a standard deviation of the first object and wherein the threshold is additionally based on the standard deviation.

29. The method of claim 25, wherein a voxel has a common attribute close to a value of the seed voxel if the voxel has an intensity within a predetermined range of the intensity of the seed voxel.

30. The method of claim 25, wherein the seed voxel is at a center of gravity of a region of the image.

31. The method of claim 25, wherein the image comprises a series of slices, each slice representing a cross section of the image.

32. The method of claim 31, wherein each slice comprises a series of frames, each frame representing a cross section of the image at a different point in time.

33. The method of claim 25, wherein the seed voxel is located within a mid-object slice and growing a test object comprises propagating the seed voxel to the remaining slices.

34. The method of claim 25, wherein each voxel in the set of voxels is weighed based on a full-blood intensity corresponding to a full-blood voxel and a myocardium intensity corresponding to a full-myocardium voxel.

35. The method of claim 25, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

36. The method of claim 25, further comprising correcting the image.

37. The method of claim 25, further comprising determining a volume of the object from the test object when its growth has stopped.

38. The method of claim 37, wherein determining a volume of the object includes computing the sum of volumes of a set of voxels weighed according to their intensity.

39. A method for locating one or more edges of an object in an image, the object bordering another object, the method comprising:

iteratively determining a volume of a test object, the test object comprising a connected set of voxels containing a seed voxel within the first object and having values of a common attribute within a range of a value of the seed voxel, the range updated for each iteration; and

computing a rate of growth of the test object between successive iterations;

stopping iteratively determining the volume of the test object at an effusion iteration when the rate of growth exceeds a predetermined value.

40. The method of claim 39, wherein the object comprises a tissue type and said another object comprises another tissue type.

41. The method of claim 39, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

42. The method of claim 39, further comprising correcting the image.

43. The method of claim 39, wherein the common attribute is based on intensity.

44. The method of claim 39, wherein the image comprises a plurality of slices and the method further comprises propagating the seed voxel to remaining slices.

45. A method for identifying a set of voxels forming an object in an image, the object bordered by another object, the method comprising:

growing a test object starting at a seed voxel within the object;

computing the volume of the test object based on a set of voxels having values of a common attribute close to a value of the seed voxel; and

stopping the growth of the test object when a rate of expansion exceeds a predetermined value.

46. The method of claim 45, wherein the object comprises a tissue type and said another object comprises another tissue type.

47. The method of claim 45, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

48. The method of claim 45, further comprising correcting the image.

49. The method of claim 45, wherein the common attribute is based on intensity.

50. The method of claim 45, wherein the image comprises a plurality of slices and wherein the method further comprises propagating the seed to remaining slices.

51. A method for measuring the volume of an object in an image comprising voxels, the object meeting other objects in the image at on or more borders, the method comprising:

identifying an attribute of the object whose value is different from values of a same or similar attribute of the other objects;

initializing a contour at one of the an borders of the object; and

deforming the contour to pass substantially through a region of voxels characteristically similar to the object, thereby forming another border and defining a space between said one of the borders and said another border that includes the region.

52. The method of claim 51, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

53. The method of claim 51, further comprising correcting the image.

54. The method of claim 51, wherein initializing a contour at an inner border comprises locating the inner border using iterative thresholding.

55. The method of claim 54, wherein iterative thresholding comprises:

growing a test object starting at a seed voxel within said another object;

computing the volume of the test object based on a set of voxels having values of a common attribute close to a value of the seed voxel, and

stopping the growth of the test object when a rate of expansion exceeds a predetermined value.

56. The method of claim 51, wherein deforming the contour includes using an active contour model incorporating image gradient vector force.

57. The method of claim 56, wherein the image gradient vector force is conditioned to causes at least one point of the contour to move in a predetermined direction when the image gradient vector force at that point is below a threshold.

58. The method of claim 56, wherein the image gradient vector force is computed using an edge detection algorithm to locate one or more edges of said another object.

59. The method of claim 55, wherein the object has an object standard deviation of its intensity and one or said other objects has another object standard deviation of its intensity and wherein a voxel has a value close to a value of the seed voxel if the voxel is within a predetermined range having an upper bound and a lower bound depending on the object standard deviation, said another object standard deviation, and the mean intensity.

60. The method of claim 51, wherein the attribute is a mean intensity of the object and wherein a voxel is characteristically similar to the object if the voxel has an intensity within a predetermined range of the mean intensity.

61. The method of claim 51, further comprising converting coordinates of the image into polar coordinates with origin at a center of mass of said another object.

62. The method of claim 61, wherein the center of mass is a center of intensity of said another object.

63. The method of claim 51, wherein deforming the contour includes minimizing an energy of the contour.

64. The method of claim 51, wherein deforming the contour includes smoothing the contour.

65. The method of claim 51, wherein deforming the contour is performed according to gradients of points on one side of the contour.

66. The method of claim 65, wherein each gradient has a direction to which the gradient influences the contour and wherein the method further comprises reversing the direction of gradients for which the direction is contrary to a predetermined direction.

67. The method of claim 51, further comprising computing a volume of the object based on the space between said one of the borders and said another border.

68. A method for locating an outer border of an object in an image comprising voxels, the object meeting other objects in the image at one or more borders, the method comprising:

identifying an attribute of the object whose value is different from values of a same or similar attribute of the other objects;

initializing a contour at one of the borders of the object; and

deforming the contour to pass substantially through a region of voxels characteristically similar to the object.

69. The method of claim 68, wherein the object comprises a tissue type and said another object comprises another tissue type.

70. The method of claim 68, wherein the data is collected from one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

71. The method of claim 68, further comprising correcting the image.

72. The method of claim 68, wherein the attribute is based on intensity values.

73. A method for identifying a set of voxels forming an object in an image, the object meeting other objects at one or more borders, the method comprising:

identifying an attribute of the objects whose value varies among the objects;

initializing a contour at one of the borders of one of said other objects; and

deforming the contour until substantially every point of the contour is adjacent to a set of voxels having an attribute whose one or more values are dissimilar to the value of the attribute of the object.

74. The method of claim 73, wherein the object comprises a tissue type and said another object comprises another tissue type.

75. The method of claim 73, wherein the data is collected from one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

76. The method of claim 73, further comprising correcting the image.

77. The method of claim 73, wherein the image comprises a series of slices and method further comprises expanding the contour among the slices.

78. The method of claim 73, wherein the attribute is based on intensity values.

79. The method of claim 73 for identifying the object and additionally for finding one or more additional objects in a nested relationship with the object, the method further comprising:

initializing an additional contour at a border of an identified object;

deforming the additional contour until substantially every point of the additional contour is adjacent to a set of voxels having an attribute whose value is dissimilar to the value of the attribute of the identified object.

80. The method of claim 79, wherein the additional contour is a deformed contour used to identify the identified object.

81. The method of claim 73, further comprising using the set of voxels forming an object in an image to identify an additional set of voxels forming one of said other objects.

82. A computer-readable medium having thereon computer-executable instructions for performing a method for measuring the volume of an object in an image comprising voxels, the object meeting other objects in the image at on or more borders, the computer-executable instructions including:

instructions for identifying an attribute of the object whose value is different from values of a same or similar attribute of the other objects;

instructions for initializing a contour at one of the an borders of the object; and

instructions for deforming the contour to pass substantially through a region of voxels characteristically similar to the object, thereby forming another border and defining a space between said one of the borders and said another border that includes the region.

83. The computer-readable medium of claim 82, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

84. The computer-readable medium of claim 82, further comprising instructions for correcting the image.

85. The computer-readable medium of claim 82, wherein initializing a contour at an inner border comprises locating the inner border using iterative thresholding.

86. The computer-readable medium of claim 85, wherein iterative thresholding comprises:

growing a test object starting at a seed voxel within said another object;

computing the volume of the test object based on a set of voxels having values of a common attribute close to a value of the seed voxel, and

stopping the growth of the test object when a rate of expansion exceeds a predetermined value.

87. The computer-readable medium of claim 82, wherein deforming the contour includes using an active contour model incorporating image gradient vector force.

88. The computer-readable medium of claim 87, wherein the image gradient vector force is conditioned to causes at least one point of the contour to move in a predetermined direction when the image gradient vector force at that point is below a threshold.

89. The computer-readable medium of claim 87, wherein the image gradient vector force is computed using an edge detection algorithm to locate one or more edges of said another object.

90. The computer-readable medium of claim 86, wherein the object has an object standard deviation of its intensity and one or said other objects has another object standard deviation of its intensity and wherein a voxel has a value close to a value of the seed voxel if the voxel is within a predetermined range having an upper bound and a lower bound depending on the object standard deviation, said another object standard deviation, and the mean intensity.

91. The computer-readable medium of claim 82, wherein the attribute is a mean intensity of the object and wherein a voxel is characteristically similar to the object if the voxel has an intensity within a predetermined range of the mean intensity.

92. The computer-readable medium of claim 82, further comprising instructions for converting coordinates of the image into polar coordinates with origin at a center of mass of said another object.

93. The computer-readable medium of claim 90, wherein the center of mass is a center of intensity of said another object.

94. The computer-readable medium of claim 82, wherein deforming the contour includes minimizing an energy of the contour.

95. The computer-readable medium of claim 82, wherein deforming the contour includes smoothing the contour.

96. The computer-readable medium of claim 82, wherein deforming the contour is performed according to gradients of points on one side of the contour.

97. The computer-readable medium of claim 82, wherein each gradient has a direction to which the gradient influences the contour and wherein the method further comprises reversing the direction of gradients for which the direction is contrary to a predetermined direction.

98. The computer-readable medium of claim 82, further comprising instructions for computing a volume of the object based on the space between said one of the borders and said another border.

99. A method for measuring the volume of an object in an image comprising voxels, the object bordering other objects in the image, the method comprising:

identifying an attribute of the object whose value is different from values of a same or similar attribute of the other objects;

initializing a contour of the object that employs the attribute;

expanding the contour using the attribute until the contour passes substantially through a region of voxels characteristically similar to the object, thereby forming first and second borders of the object and defining a space between the first and second borders that includes the region.

100. The method of claim 99, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

101. The method of claim 99, further comprising correcting the image.

102. The method of claim 99, wherein the contour is initialized at the first border of the object and the expansion proceeds on a single side of the contour.

103. The method of claim 99, wherein the contour is initialized centrally within the object and the expansion occurs on both sides of the contour.

104. The method of claim 103, wherein the contour is initialized by searching a graph having nodes comprising image data.

105. The method of claim 104, wherein the graph is constructed using an estimation of first and second edges of the object.

106. The method of claim 104, further comprising assigning a cost to each node and wherein searching the graph includes locating a minimal cost path.

107. The method of claim 99, further comprising computing a volume of the object based on the space between the first and second borders.

108. The method of claim 99, wherein the object is a heart tissue and one of said said other objects is blood contained by the heart tissue.

109. The method of claim 99, wherein the voxels comprise imaging data collected from one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

110. The method of claim 102, wherein initializing a contour at the first border comprises locating the first border using iterative thresholding.

111. The method of claim 110, wherein iterative thresholding comprises:

growing a test object starting at one or more seed voxels within one of said other objects;

computing the volume of the test object based on a set of voxels having values of a common attribute close to a value of the seed voxel, and

stopping the growth of the test object when a rate of expansion exceeds a predetermined value.

112. The method of claim 102, wherein the expansion of the contour includes using an active contour model incorporating an image gradient vector force.

113. The method of claim 112, wherein the image gradient vector force causes a point of the contour to move in a predetermined direction when the image gradient vector force at that point is below a threshold.

114. The method of claim 112, wherein the image gradient vector force is computed using an edge detection algorithm to locate one or more edges of one of said other objects.

115. The method of claim 99, wherein the attribute is a mean intensity of the object and wherein a voxel is characteristically similar to the object if the voxel has an intensity within a predetermined range of the mean intensity.

116. The method of claim 115, wherein the object has an object standard deviation of its intensity and one of said other objects has another object standard deviation of its intensity and wherein the predetermined range has upper and lower bounds depending on the object standard deviation, said another object standard deviation, and the mean intensity.

117. The method of claim 99, further comprising converting coordinates of the image into polar coordinates with an origin proximate to a center of mass of one of said other objects.

118. The method of claim 117, wherein the center of mass is a center of intensity of one of said other objects.

119. The method of claim 99, wherein the expansion of the contour includes at least one of (1) minimizing an energy of the contour and (2) smoothing the contour.

120. The method of claim 99, wherein the expansion of the contour is performed according to gradients of points around the contour.

121. The method of claim 99, wherein the expansion of the contour is performed according to gradients of points on one side of the contour.

122. The method of claim 120, wherein each gradient has a direction to which the gradient influences the contour and wherein the method further comprises reversing the direction of gradients opposing an intended direction of expansion.

123. A method for identifying an attribute of an object represented by imaging data, where the object contains a volume, the method comprising:

determining inner and outer edges of the object using one or more values of one or more attributes of the imaging data comprising the contained volume;

extracting the object using the inner and outer edges of the object;

determining a value of the volume for the object by growing the volume from a hollow contained by the object to the extracted object.

124. The method of claim 123, wherein the object is a myocardium and the contained volume comprises blood.

125. The method of claim 123, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

126. The method of claim 123, further comprising correcting the image.

127. The method of claim 123 wherein the determination of the inner and outer edges of the object includes extracting edge information from a circular map of the imaging data using an edge detection algorithm.

128. The method of claim 123, wherein the one or more values of one or more attributes includes a value of an intensity of the image data comprising the contained volume.

129. The method of claim 123, including determining an initial value of the volume and the value of the intensity is a mean value taken from the initial value of the volume.

130. The method of claim 124, wherein the final value of the volume is used to measure an ejection fraction.

131. The method of claim 124, wherein the extracted object is used to measure a mass of the myocardium.

132. The method of claim 123, wherein the extraction of the object includes finding a center axis of the object in the imaging data and expanding the imaging data comprising the object using graph searching.

133. A method of resolving first and second edges of an object in imaging data, the method comprising,

determining characteristic values of at least one attribute of areas of the imaging data inside and outside of the object;

searching for changes in the value of the attribute that mark an edge of the object;

generating a histogram between first and second edges of the object in the imaging data; and

extracting the object from the imaging data based on values of the attribute taken from the histogram.

134. The method of claim 133, wherein the imaging data is converted to polar coordinates prior to searching for changes in the value of the attribute and converting the image to Cartesian coordinates after generating the histogram.

135. The method of claim 133, wherein the object is heart tissue.

136. The method of claim 133, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

137. The method of claim 133, further comprising correcting the image.

138. A method for identifying a volume bordered by an object at a boundary, the object and volume represented by imaging data, the method comprising:

initializing a contour within the object;

detecting the boundary of the object by growing a first region from the contour;

expanding a second region in one or more directions from within the volume;

stopping expansion of the second region in a direction when growth in the direction comes in contact with the first region of the object; and

determining a value of the volume from the second region after the second region stops expanding in each of the one or more directions.

139. The method of claim 138, wherein the object is heart tissue and the second region comprises blood contained within a hollow of the heart tissue.

140. The method of claim 138, wherein the image is obtained using one or more of (1) magnetic resonance imaging, (2) computed tomography (CT), (3) ultrasound, (4) X-ray and (5) single photon emission computed tomography (SPECT).

141. The method of claim 138, further comprising correcting the image.

142. The method of claim 138, wherein initializing a contour within the object includes constructing a graph from the imaging data and locating the contour based on results of a search on the graph.

143. The method of claim 138, further comprising detecting one or more discontinuities in the imaging data in order to identify one or more characteristics of the object, wherein the initialization of the contour within the object includes locating the contour according to the one or more characteristics.