Massive training artificial neural network (MTANN) for detecting abnormalities in medical images

• US 6,819,790 B2
• Filed: 04/12/2002
• Issued: 11/16/2004
• Est. Priority Date: 04/12/2002
• Status: Expired due to Term

• Current Assignee: University of Chicago The (University of Chicago)
• Original Assignee: University of Chicago
• Inventors: Suzuki, Kenji, Doi, Kunio
• Primary Examiner(s): Do, Anh Hong
• Application Number:
  US10/120,420
• Publication Number:
  US 20030194124A1
• Time in Patent Office: 949 Days
• Field of Search: 382/156, 382/157, 382/118, 382/261, 382/130, 382/131, 600/300
• US Class Current: 382/156
• CPC Class Codes: G06T 7/0012 Biomedical image inspection

Abstract

A method of training an artificial neural network (ANN) involves receiving a likelihood distribution map as a teacher image, receiving a training image, moving a local window across sub-regions of the training image to obtain respective sub-region pixel sets, inputting the sub-region pixel sets to the ANN so that it provides output pixel values that are compared to output pixel values of corresponding teacher image pixel values to determine an error, and training the ANN to reduce the error. A method of detecting a target structure in an image involves scanning a local window across sub-regions of the image by moving the local window for each sub-region so as to obtain respective sub-region pixel sets, inputting the sub-region pixel sets to an ANN so that it provides respective output pixel values that represent likelihoods that respective image pixels are part of a target structure, the output pixel values collectively constituting a likelihood distribution map. Another method for detecting a target structure involves training N parallel ANNs on either (A) a same target structure and N mutually different non-target structures, or (B) a same non-target structure and N mutually different target structures, the ANNs outputting N respective indications of whether the image includes a target structure or a non-target structure, and combining the N indications to form a combined indication of whether the image includes a target structure or a non-target structure. The invention provides related apparatus and computer program products storing executable instructions to perform the methods.

Claims

1. A method of training an artificial neural network including network parameters that govern how the artificial neural network operates the method comprising:

receiving at least a training image including plural training image pixels;

receiving at least a likelihood distribution map as a teacher image, the teacher image including plural teacher image pixels each having a pixel value indicating likelihood that a respective training image pixel is part of a target structure;

moving a local window across plural sub-regions of the training image to obtain respective sub-region pixel sets;

inputting the sub-region pixel sets to the artificial neural network so that the artificial neural network provides output pixel values;

comparing the output pixel values to corresponding teacher image pixel values to determine an error; and

training the network parameters of the artificial neural network to reduce the error.

2. The method of claim 1, wherein:

the training image receiving step includes receiving at least a training medical image; and

the local window moving step includes moving the local window across the plural sub-regions of the training medical image to obtain the respective sub-region pixel sets.

3. The method of claim 2, wherein the step of receiving at least a training medical image includes:

receiving one or more training medical images that include an abnormality and a normal anatomical structure.

4. The method of claim 3, further comprising:

coordinating the step of receiving at least an abnormality likelihood distribution map and the step of receiving at least a training medical image, so that;

(1) when the training medical image includes an abnormality, the teacher image pixel values represent likelihoods that corresponding training medical image pixels are part of an abnormality; and

(2) when the training medical image does not include an abnormality, the teacher image pixel values represent an absence of an abnormality at corresponding training medical image pixel locations.

5. The method of claim 4, wherein, when the training medical image includes an abnormality, the step of receiving the teacher image includes:

receiving a Gaussian distribution map whose pixels have respective pixel values that represent a likelihood that the pixel is part of an abnormality.

6. The method of claim 5, wherein the step of receiving a Gaussian distribution map includes:

receiving the Gaussian distribution map having a standard deviation proportional to a size of the abnormality.

7. The method of claim 2, wherein the local window moving step includes:

scanning the local window across consecutively physically overlapping sub-regions of the training medical image by moving the local window a predetermined distance for each sub-region.

8. The method of claim 7, wherein the local window moving step includes:

scanning the local window across the consecutively physically overlapping sub-regions of the training medical image by moving the local window a predetermined distance equal to a pixel pitch value in the training medical image, so that successive sub-regions are offset from each other by a separation distance of adjacent pixels in the training medical image.

9. A method of detecting an abnormality in a medical image by using an artificial neural network, the method comprising:

training the artificial neural network using the method of claim 2;

scanning a local window across consecutively physically overlapping sub-regions of the medical image by moving the local window a predetermined distance for each sub-region, so as to obtain respective sub-region pixel sets;

inputting the sub-region pixel sets into the artificial neural network so that the artificial neural network provides, corresponding to the sub-regions, respective output pixel values that each represent a likelihood that respective medical image pixels are part of an abnormality, the output pixel values collectively constituting a likelihood distribution map; and

scoring the likelihood distribution map to detect the abnormality.

10. An artificial neural network including network parameters that govern how the artificial neural network operates, the artificial neural network being trained by the steps of:

receiving at least a training image including plural training image pixels;

receiving at least a likelihood distribution map as a teacher image, the teacher image including plural teacher image pixels each having a pixel value indicating a likelihood that a respective training image pixel is part of a target structure;

moving a local window across plural sub-regions of the training image to obtain respective sub-region pixel sets;

inputting the sub-region pixel sets to the artificial neural network so that the artificial neural network provides output pixel values;

comparing the output pixel values to corresponding teacher image pixel values to determine an error; and

training the network parameters of the artificial neural network to reduce the error.

11. The artificial neural network of claim 10, wherein:

the training image receiving step includes receiving at least a training medical image; and

the local window moving step includes moving the local window across the plural sub-regions of the training medical image to obtain the respective sub-region pixel sets.

12. The artificial neural network of claim 11, wherein the step of receiving at least a training medical image includes:

receiving one or more training medical images that include an abnormality and a normal anatomical structure.

13. The artificial neural network of claim 12, trained by the additional steps of:

coordinating the step of receiving at least an abnormality likelihood distribution map and the step of receiving at least a training medical image, so that;

(1) when the training medical image includes an abnormality, the teacher image pixel values represent likelihoods that corresponding training medical image pixels are part of an abnormality; and

(2) when the training medical image does not include an abnormality, the teacher image pixel values represent an absence of an abnormality at corresponding training medical image pixel locations.

14. The artificial neural network of claim 13, wherein, when the training medical image includes an abnormality, the step of receiving the teacher image includes:

receiving a Gaussian distribution map whose pixels have respective pixel values that represent a likelihood that the pixel is part of an abnormality.

15. The artificial neural network of claim 14, wherein the step of receiving a Gaussian distribution map includes:

receiving the Gaussian distribution map having a standard deviation proportional to a size of the abnormality.

16. The artificial neural network of claim 11, wherein the local window moving step includes:

scanning the local window across consecutively physically overlapping sub-regions of the training medical image by moving the local window a predetermined distance for each sub-region.

17. The artificial neural network of claim 16, wherein the local window moving step includes:

scanning the local window across the consecutively physically overlapping sub-regions of the training medical image by moving the local window a predetermined distance equal to a pixel pitch value in the training medical image, so that successive sub-regions are offset from each other by a separation distance of adjacent pixels in the training medical image.

18. A computer program product storing program instructions for execution on a computer system, which when executed by the computer system, cause the computer system to train an artificial neural network having network parameters that govern how the artificial neural network operates by performing the steps of:

receiving at least a training image including plural training image pixels;

receiving at least a likelihood distribution map as a teacher image, the teacher image including plural teacher image pixels each having a pixel value indicating a likelihood that a respective training image pixel is part of a target structure;

moving a local window across plural sub-regions of the training image to obtain respective sub-region pixel sets;

inputting the sub-region pixel sets to the artificial neural network so that the artificial neural network provides output pixel values;

comparing the output pixel values to corresponding teacher image pixel values to determine an error; and

training the network parameters of the artificial neural network to reduce the error.

19. The computer program product of claim 18, wherein:

the training image receiving step includes receiving at least a training medical image; and

the local window moving step includes moving the local window a across the plural sub-regions of the training medical image to obtain the respective sub-region pixel sets.

20. The computer program product of claim 19, wherein the step of receiving at least a training medical image includes:

receiving one or more training medical images that include an abnormality and a normal anatomical structure.

21. The computer program product of claim 20 storing further program instructions for execution on the computer system, which when executed by the computer system, cause the computer system to train an artificial neural network having network parameters that govern how the artificial neural network operates by performing the additional steps of:

coordinating the step of receiving at least an abnormality likelihood distribution map and the step of receiving at least a training medical image, so that;

(1) when the training medical image includes an abnormality, the teacher image pixel values represent likelihoods that corresponding training medical image pixels are part of an abnormality; and

(2) when the training medical image does not include an abnormality, the teacher image pixel values represent an absence of an abnormality at corresponding training medical image pixel locations.

22. The computer program product of claim 21, wherein, when the training medical image includes an abnormality, the step of receiving the teacher image includes:

receiving a Gaussian distribution map whose pixels have respective pixel values that represent a likelihood that the pixel is part of an abnormality.

23. The computer program product of claim 22, wherein the step of receiving a Gaussian distribution map includes:

receiving the Gaussian distribution map having a standard deviation proportional to a size of the abnormality.

24. The computer program product of claim 19, wherein the local window moving step includes:

scanning the local window across consecutively physically overlapping sub-regions of the training medical image by moving the local window a predetermined distance for each sub-region.

25. The computer program product of claim 24, wherein the local window moving step includes:

scanning the local window across the consecutively physically overlapping sub-regions of the training medical image by moving the local window a predetermined distance equal to a pixel pitch value in the training medical image, so that successive sub-regions are offset from each other by a separation distance of adjacent pixels in the training medical image.

26. A method of detecting a target structure in an image by using an artificial neural network, the method comprising:

scanning a local window across sub-regions of the image by moving the local window for each sub-region, so as to obtain respective sub-region pixel sets;

inputting the sub-region pixel sets to the artificial neural network so that the artificial neural network provides, corresponding to the sub-regions, respective output pixel values that represent likelihoods that respective image pixels are part of a target structure, the output pixel values collectively constituting a likelihood distribution map; and

scoring the likelihood distribution map to detect the target structure.

27. The method of claim 26, wherein:

the image is a medical image;

the target structure is an abnormality in the medical image;

the scanning step includes scanning the local window across sub-regions of the medical image;

the artificial neural network provides the respective output pixel values that represent the likelihoods that the respective medical image pixels are part of an abnormality; and

the scoring step includes scoring the likelihood distribution map to detect the abnormality.

28. The method of claim 27, wherein the scanning step includes:

scanning the local window across consecutively physically overlapping sub-regions of the medical image by moving the local window a predetermined distance for each sub-region, so as to obtain respective sub-region pixel sets.

29. The method of claim 28, wherein:

the predetermined distance is a pixel pitch value in the medical image, so that successive sub-regions are offset from each other by a separation distance of adjacent pixels in the medical image.

30. The method of claim 28, wherein the scoring step includes:

filtering the output pixel values in the likelihood distribution map.

31. The method of claim 30, further comprising:

comparing results of the filtering to at least a threshold value to detect the abnormality.

32. The method of claim 30, wherein the filtering step includes:

filtering the output pixel values with a Gaussian function.

33. The method of claim 28, wherein the artificial neural network includes:

an output layer including units having linear activation functions.

34. An artificial neural network system, comprising:

an artificial neural network configured to detect a target structure in an image;

a scanning mechanism configured to scan a local window across sub-regions of the image by moving the local window for each sub-region, so as to obtain respective sub-region pixel sets;

an inputting mechanism configured to input the sub-region pixel sets to the artificial neural network so that the artificial neural network provides, corresponding to the sub-regions, respective output pixel values that represent likelihoods that respective image pixels are part of the target structure, the output pixel values collectively constituting a likelihood distribution map; and

a scoring mechanism configured to score the likelihood distribution map to detect the target structure.

35. The artificial neural network system of claim 34, wherein:

the image is a medical image;

the target structure is an abnormality in the medical image;

the mechanism for scanning includes means for scanning the local window across sub-regions of the medical image;

the artificial neural network is configured to output the respective output pixel values that represent the likelihoods that the respective medical image pixels are part of an abnormality; and

the mechanism for scoring includes means for scoring the likelihood distribution map to detect the abnormality.

36. The artificial neural network system of claim 35, wherein the mechanism for scanning includes:

means for scanning the local window across consecutively physically overlapping sub-regions of the medical image by moving the local window a predetermined distance for each sub-region, so as to obtain respective sub-region pixel sets.

37. The artificial neural network system of claim 36, wherein:

the predetermined distance is a pixel pitch value in the medical image, so that successive sub-regions are offset from each other by a separation distance of adjacent pixels in the medical image.

38. The artificial neural network system of claim 36, wherein the means for scoring includes:

a filter configured to filter the output pixel values in the likelihood distribution map.

39. The artificial neural network system of claim 38, further comprising:

a comparing mechanism for comparing results of the means for filtering to at least a threshold value to detect the abnormality.

40. The artificial neural network system of claim 38, wherein the filter is configured to filter the output pixel values with a Gaussian function.

41. The artificial neural network system of claim 36, wherein the artificial neural network includes:

an output layer including units having linear activation functions.

42. A computer program product storing program instructions for execution on a computer system, which when executed by the computer system, cause the computer system to detect a target structure in an image by using an artificial neural network by performing the steps of:

scanning a local window across sub-regions of the image by moving the local window for each sub-region, so as to obtain respective sub-region pixel sets;

inputting the sub-region pixel sets to the artificial neural network so that the artificial neural network provides corresponding to the sub-regions respective output pixel values that represent likelihoods that respective image pixels are part of a target structure, the output pixel values collectively constituting a likelihood distribution map; and

scoring the likelihood distribution map to detect the target structure.

43. The computer program product of claim 42, wherein:

the image is a medical image;

the target structure is an abnormality in the medical image;

the scanning step includes scanning the local window across sub-regions of the medical image;

the artificial neural network provides the respective output pixel values that represent the likelihoods that the respective medical image pixels are part of an abnormality; and

the scoring step includes scoring the likelihood distribution map to detect the abnormality.

44. The computer program product of claim 43, wherein the scanning step includes:

scanning the local window across consecutively physically overlapping sub-regions of the medical image by moving the local window a predetermined distance for each sub-region, so as to obtain respective sub-region pixel sets.

45. The computer program product of claim 44, wherein:

the predetermined distance is a pixel pitch value in the medical image, so that successive sub-regions are offset from each other by a separation distance of adjacent pixels in the medical image.

46. The computer program product of claim 44, wherein the scoring step includes:

filtering the output pixel values in the likelihood distribution map.

47. The computer program product of claim 46 storing program instructions for execution on the computer system, which when executed by the computer system, cause the computer system to detect a target structure in an image by using an a artificial neural network by performing the additional steps of:

comparing results of the filtering to at least a threshold value to detect the abnormality.

48. The computer program product of claim 46, wherein the filtering step includes:

filtering the output pixel values with a Gaussian function.

49. The computer program product of claim 44, wherein the artificial neural network includes:

an output layer including units having linear activation functions.

50. An apparatus for detecting a target structure in an image, the apparatus comprising:

a network configured to receive sub-region pixel sets from respective sub-regions of the image, and to operate on the sub-region pixel sets so as to produce a likelihood distribution map including output pixel values that represent likelihoods that corresponding image pixels are part of the target structure.

51. The apparatus of claim 50, wherein:

the image is a medical image;

the target structure is an abnormality in the medical image; and

the network is configured to receive the sub-region pixel sets from the respective sub-regions of the medical image, and to operate on the sub-region pixel sets so as to produce the likelihood distribution map including the output pixel values that represent the likelihoods that corresponding medical image pixels are part of the abnormality.

52. The apparatus of claim 51, comprising:

a pixel set generation mechanism configured to apply to the network sub-region pixel sets from respective consecutively physically overlapping sub-regions of the medical image.

53. The apparatus of claim 52, wherein the network operates in accordance with network parameters, and the apparatus further comprises:

a training portion configured to train the network parameters in accordance with (a) teacher image pixel values that represent a likelihood distribution of pixels portraying a detected abnormality, and (b) a training period likelihood distribution map that includes output pixels produced by the network operating on at least one training medical image, the output pixels respectively representing likelihoods that corresponding medical image pixels are part of the abnormality.

54. The apparatus of claim 53, wherein:

the network is further configured to receive, during a training period, sub-region pixel sets from respective consecutively physically overlapping sub-regions of the at least one training medical image, and to operate in accordance with the network parameters on the sub-region pixel sets so as to produce the training period likelihood distribution map.

55. An apparatus for detecting an abnormality in a medical image, the apparatus comprising:

first through N-th artificial neural networks constituting respective apparatus formed in accordance with any of claims 52, 53 or 54, N being an integer greater than 1, that have been commonly trained on a same abnormality and on first through N-th mutually different normal structures, the first through N-th artificial neural networks being configured to output first through N-th respective indications of whether a structure in the medical image is an abnormality or a normal anatomical structure; and

a combiner that is configured to combine the first through N-th indications to form a combined indication of whether the structure in the medical image is an abnormality or a normal anatomical structure.

56. A method for detecting an abnormality in a medical image, the method comprising:

training first through N-th artificial neural networks constituting respective apparatus formed in accordance with any of claims 52, 53 or 54, N being an integer greater than 1, on a same abnormality and on first through N-th mutually different normal an anatomical structures, the first through N-th artificial neural networks being configured to output first through N-th respective indications of whether the medical image includes an abnormality or a normal anatomical structure; and

combining the first through N-th indications to form a combined indication of whether the medical image includes an abnormality or a normal anatomical structure.

57. The apparatus of any of claims 52, 53 or 54, wherein:

the network constitutes an artificial neural network.

58. The apparatus of claim 57, wherein the artificial neural network includes:

an output layer including at least one unit having a linear activation function.

59. The apparatus of either of claims 53 or 54, wherein the likelihood distribution constitutes:

a Gaussian function centered about a center of a portrayal of a detected abnormality.

60. The apparatus of claim 59, wherein:

a standard deviation of the Gaussian function varies with a size of the portrayal of the detected abnormality.

61. The apparatus of claim 52, further comprising:

a filter configured to receive the likelihood distribution map from the network and to provide a score whose value indicates whether the medical image includes an abnormality.

62. The apparatus of claim 52, wherein the pixel set generation mechanism comprises:

a scanning mechanism configured to scan a local window across an input image by moving the local window a predetermined distance for each sub-region.

63. The apparatus of claim 62, wherein:

the predetermined distance is a pixel pitch value in the input image, so that successive sub-regions are offset from each other by a separation distance of adjacent pixels in the input image.

64. A method for detecting a target structure in an image, the method comprising:

training first through N-th artificial neural networks, N being an integer greater than 1, on either (A) a same target structure and first through N-th mutually different non-target structures, or (B) a same non-target structure and first through N-th mutually different target structures, the first through N-th artificial neural networks being configured to output first through N-th respective indications of whether the image includes a target structure or a non-target structure; and

combining the first through N-th indications to form a combined indication of whether the image includes a target structure or a non-target structure.

65. The method of claim 64, wherein:

the image is a medical image;

the target structure is an abnormality in the medical image;

the non-target structures are normal anatomical structures;

the training step includes training the first through N-th artificial neural networks on either (A) a same abnormality and first through N-th mutually different normal anatomical structures, or (B) a same normal anatomical structure and first through N-th mutually different abnormalities, the first through N-th artificial neural networks being configured to output first through N-th respective indications of whether the medical image includes an abnormality or a normal anatomical structure; and

the combining step includes combining the first through N-th indications to form a combined indication of whether the medical image includes an abnormality or a normal anatomical structure.

66. The method of claim 65, wherein the training step includes:

training first through N-th artificial neural networks on the same abnormality and on the first through N-th mutually different normal anatomical structures.

67. The method of claim 66, wherein the combining step includes:

providing a combined indication of an abnormality using a logical AND combiner.

68. The method of claim 67, wherein the combining step includes:

providing a combined indication of an abnormality, only if all first through N-th respective indications indicate an abnormality.

69. The method of claim 65, wherein:

the training step includes training first through N-th artificial neural networks on the same normal anatomical structure and on the first through N-th mutually different abnormalities; and

the combining step includes providing a combined indication of a normal anatomical structure, only if all first through N-th respective indications indicate a normal anatomical structure.

70. The method of claim 65, wherein the training step includes training the artificial neural networks using a normal anatomical structure include at least one from a group including:

large vessels in the hilum;

large vessels with opacities;

medium-sized vessels;

small vessels;

soft-tissue opacities caused by a heart;

soft-tissue opacities caused by a diaphragm;

soft-tissue opacities caused by a partial volume effect between peripheral vessels and the diaphragm;

abnormal opacities; and

focal infiltrative opacities.

71. A computer program product storing program instructions for execution on a computer system, which when executed by the computer system, cause the computer system to detect a target structure in an image by performing the steps of:

training first through N-th artificial neural networks, N being an integer greater than 1, on either (A) a same target structure and first through N-th mutually different non-target structures, or (B) a same non-target structure and first through N-th mutually different target structures, the first through N-th artificial neural networks being configured to output first through N-th respective indications of whether the image includes a target structure or a non-target structure; and

combining the first through N-th indications to form a combined in indication of whether the image includes a target structure or a non-target structure.

72. The computer program product of claim 71, wherein:

the image is a medical image;

the target structure is an abnormality in the medical image;

the non-target structures are normal anatomical structures;

the training step includes training the first through N-th artificial neural networks on either (A) a same abnormality and first through N-th mutually different normal anatomical structures, or (B) a same normal anatomical structure and first through N-th mutually different abnormalities, the first through N-th artificial neural networks being configured to output first through N-th respective indications of whether the medical image includes an abnormality or a normal anatomical structure; and

the combining step includes combining the first through N-th indications to form a combined indication of whether the medical image includes an abnormality or a normal anatomical structure.

73. The computer program product of claim 72, wherein the training step includes:

first through N-th artificial neural networks on the same abnormality and on the first through N-th mutually different normal anatomical structures.

74. The computer program product of claim 73, wherein the combining step includes:

providing a combined indication of an abnormality using a logical AND combiner.

75. The computer program product of claim 74, wherein the combining step includes:

providing a combined indication of an abnormality, only if all first through N-th respective indications indicate an abnormality.

76. The computer program product of claim 72, wherein:

the training step includes training first through N-th artificial neural networks on the same normal anatomical structure and on the first through N-th mutually different abnormalities; and

the combining step includes providing a combined indication of abnormal anatomical structure, only if all first through N-th respective indications indicate a normal anatomical structure.

77. The computer program product of claim 72, wherein he training step includes training the artificial neural networks using a normal anatomical structure include at least one from a group including:

large vessels in the hilum;

large vessels with opacities;

medium-sized vessels;

small vessels;

soft-tissue opacities caused by a heart;

soft-tissue opacities caused by a diaphragm;

soft-tissue opacities caused by a partial volume effect between peripheral vessels and the diaphragm;

abnormal opacities; and

focal infiltrative opacities.

78. An apparatus for detecting a target structure in an image, the apparatus comprising:

first through N-th artificial neural networks, N being an integer greater than 1, that have been trained on either (A) a same target structure and first through N-th mutually different non-target structures, or (B) a same non-target structure and first through N-th mutually different target structures, the first through N-th artificial neural networks being configured to output first through N-th respective indications of whether the image includes a target structure or a non-target structure; and

a combiner configured to combine the first through N-th indications to form a combined indication of whether the medical image includes a target structure or a non-target structure.

79. The apparatus of claim 78, wherein:

the image is a medical image;

the target structure is an abnormality in the medical image;

the non-target structures are normal anatomical structures;

the first through N-th artificial neural networks have been trained on either (A) a same abnormality and first through N-th mutually different normal anatomical structures, or (B) a same normal anatomical structure and first through N-th mutually different abnormalities, the first through N-th artificial neural networks being configured to output first through N-th respective indications of whether the medical image includes an abnormality or a normal anatomical structure; and

the combiner is configured to combine the first through N-th indications to form the combined indication of whether the medical image includes an abnormality or a normal anatomical structure.

80. The apparatus of claim 79, wherein:

the artificial neural networks have been trained on a same abnormality and first through N-th mutually different normal anatomical structures.

81. The apparatus of claim 80, wherein the combiner includes:

a logical AND combiner that provides a combined indication of an abnormality.

82. The apparatus of claim 81, wherein the combiner includes:

a logical AND combiner that provides a combined indication of an abnormality, only if all first through N-th respective indications indicate an abnormality.

83. The apparatus of claim 79, herein:

the artificial neural networks have been trained on a same normal anatomical structure and first through N-th mutually different abnormalities; and

the combiner includes a logical AND combiner that provides a combined indication of a normal anatomical structure, only if all first through N-th respective indications indicate a normal anatomical structure.

84. The apparatus of claim 79, wherein the artificial neural networks have been trained on a normal anatomical structure include at least one from a group including:

large vessels in the hilum;

large vessels with opacities;

medium-sized vessels;

small vessels;

soft-tissue opacities caused by a heart;

soft-tissue opacities caused by a diaphragm;

soft-tissue opacities caused by a partial volume effect between peripheral vessels and the diaphragm;

abnormal opacities; and

focal infiltrative opacities.