Method and system for the segmentation of lung regions in lateral chest radiographs

• US 6,335,980 B1
• Filed: 12/23/1999
• Issued: 01/01/2002
• Est. Priority Date: 07/25/1997
• Status: Expired due to Fees

• Current Assignee: ARCH Development Corporation (University of Chicago)
• Original Assignee: ARCH Development Corporation (University of Chicago)
• Inventors: Macmahon, Heber, Giger, Maryellen L., Armato, Samuel G. III
• Primary Examiner(s): Patel, Jayanti K.
• Application Number:
  US09/471,088
• Time in Patent Office: 740 Days
• Field of Search: 382/100, 382/128, 382/132, 382/169, 382/173, 382/260, 382/256, 382/266, 378/37, 378/8, 600/425, 600/443
• US Class Current: 382/132
• CPC Class Codes: G06T 2207/10116 X-ray image
• CPC Class Codes: G06T 2207/20012 Locally adaptive
• CPC Class Codes: G06T 2207/30061 Lung
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/136 involving thresholding
• CPC Class Codes: G06V 10/28 Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns

Abstract

A method and system for the automated segmentation of the lung regions in lateral chest radiographs. This is achieved according to the invention by providing an improved computerized, automated method for image segmentation based on gray-level threshold analysis. A unique method for identifying an approximate outer bounds on the extent of the lung fields in the image is performed to restrict the region further analyzed. An iterative global gray-level thresholding method is applied based on the features of a global gray-level histogram. Features of the regions in a binary image constructed at each iteration are identified and subjected to a modified analysis to exclude regions external to the lung field. The initial lung region contour that results from this global process is used to facilitate a novel adaptive local gray level thresholding method. Individual regions-of-interest (ROIs) are placed along the initial contour. The dimensions of the several ROIs are based upon the patient anatomy enclosed therein. A unique procedure is implemented to determine the single gray-level threshold to be applied to the pixels within the individual ROIs. A composite binary image results, and a final contour is constructed to enclose “on” regions thereof. Smoothing processes are applied, including a unique adaptation of a rolling ball method, and fitted polynomial curves are spliced into the final contour.

Claims

1. A method for the automated segmentation of the lung region in lateral chest radiographic images, comprising:

obtaining first image data representing the thorax of a laterally positioned patient;

pre-processing said first image data to produce second image data;

delineating approximate anterior and posterior margins in said second image data to produce third image data,

performing iterative global gray-level thresholding on said third image data to identify a first initial lung segmentation contour; and

smoothing said first initial lung segmentation contour to produce a second initial lung segmentation contour.

2. The method according to claim 1, further comprising:

performing adaptive local gray-level thresholding within regions-of-interest positioned along said second initial lung segmentation contour to identify a first final lung segmentation contour; and

smoothing said first final lung segmentation contour to produce a second final lung segmentation contour.

3. The method according to claim 2, wherein said performing adaptive local gray-level thresholding step comprises:

placing N local regions-of-interest of a predetermined dimension centered at points along said second initial lung segmentation contour separated by a predetermined constant number of points;

classifying each of said N local regions-of-interest as one of M anatomic location categories;

adjusting said predetermined dimension of each of said N local regions-of-interest depending on a corresponding of said M anatomic location categories; and

creating a composite binary image.

4. The method according to claim 3, wherein said creating step comprises:

calculating N local gray-level thresholds for said N local regions-of-interest depending on the corresponding of said M anatomic location categories; and

turning “

on”

pixels in said composite binary image that have corresponding pixels in each of said N local regions-of-interest in said third image data with values exceeding the corresponding of said N local gray-level thresholds.

5. The method according to claim 4, wherein said performing adaptive local gray-level thresholding step further comprises:

constructing said first final lung segmentation contour around the region of contiguous “

on”

pixels in said composite binary image.

6. The method according to claim 2, wherein said smoothing said first final lung segmentation contour step comprises:

applying a rolling ball filter to said first final lung segmentation contour to produce said second final lung segmentation contour.

7. The method according to claim 6, wherein said applying step comprises:

constructing a circular filter of a predetermined constant value in radius;

calculating a slope of a line tangent to each point of said first final lung segmentation contour;

placing a contact point found on the perimeter of the circular filter successively on each point of said first final lung segmentation contour such that the slope of a line tangent to the circle perimeter of said circular filter at said contact point matches said slope of said line tangent to each point of said first final lung segmentation contour; and

applying linear interpolation to bridge concavities identified when said ball filter contacts said first final lung segmentation contour at more than one point.

8. The method according to claim 2, wherein said smoothing said first final lung segmentation contour step further comprises:

performing a running mean operation that replaces the position of each point of said first final lung segmentation contour with the average position of a predetermined constant number of adjacent contour points to construct a third final lung segmentation contour; and

eliminating points in said third final lung segmentation contour that are redundant in an eight-point connectivity sense to construct said second final lung segmentation contour.

9. The method according to claim 2, wherein said smoothing said first final lung segmentation contour step further comprises:

fitting separate polynomial functions to an anterior aspect of said first final lung segmentation contour and to a posterior aspect of said first final lung segmentation contour.

10. The method according to claim 9, wherein said fitting step comprises:

identifying as anterior aspect points the points along said first final lung segmentation contour from a superiormost point of said first final lung segmentation contour counterclockwise to the point along said first final lung segmentation contour possessing the smallest geometric distance to the bottom anterior corner of said second image data;

fitting an anterior least-squares polynomial function to said anterior aspect points; and

substituting said anterior aspect points with points derived from said anterior least-squares polynomial function between a superiormost anterior point of intersection and an inferiormost point of intersection of said first final lung segmentation contour and said anterior least-squares polynomial function;

identifying as posterior aspect points the points along said first final lung segmentation contour from a superiormost point of said first final lung segmentation contour clockwise to the point along said first final lung segmentation contour possessing the smallest geometric distance to the bottom posterior corner of said second image data; and

fitting a posterior least-squares polynomial function to said posterior aspect points; and

substituting said posterior aspect points with points derived from said posterior least-squares polynomial function between a superiormost point of intersection and an inferiormost point of intersection of said first final lung segmentation contour and said posterior least-squares polynomial function, thereby constructing said second final lung segmentation contour.

11. The method according to claim 1, wherein said pre-processing step comprises:

identifying bottom collimation regions of said first image data;

identifying side collimation regions of said first image data; and

setting to zero the values of pixels in said collimation regions.

12. The method according to claim 11, wherein said step of identifying bottom collimation regions comprises:

comparing the initial and final pixel values of a bottom row of said first image data to a first predetermined constant value; and

setting to zero the values of all pixels within said bottom row of said first image data when the value of said initial pixel or the value of said final pixel in said bottom row of said first image data exceeds said first predetermined constant value.

13. The method according to claim 12, further comprising:

(a) repeating said comparing and setting to zero steps for preceding rows until a first row is encountered in which neither the value of said initial pixel nor the value of said final pixel in said first row exceeds said first predetermined constant value;

(b) comparing the initial and final pixel values of a row preceding said first row of said first image data to a second predetermined constant value; and

(c) setting to zero the values of all pixels within said row preceding said first row of said first image data when the value of said initial or final pixel in said row preceding said first row of said first image data exceeds said second predetermined constant value.

14. The method according to claim 13, comprising:

repeating said steps (b) and (c) for preceding rows until a second row is encountered in which neither the value of an initial pixel nor the value of a final pixel in said second row exceeds said second predetermined constant value.

15. The method according to claim 11, wherein said step of identifying side collimation regions comprises:

comparing the value of an initial pixel in each row of said first image data with a predetermined constant value; and

setting to zero the value of said initial pixel when said initial pixel exceeds said predetermined constant value.

16. The method according to claim 15, comprising:

(a) repeating said comparing and setting to zero steps for subsequent pixels until a first pixel is encountered that does not exceed said predetermined constant value;

(b) comparing the value of a final pixel in each row of said first image data with said predetermined constant value; and

(c) setting to zero the value of said final pixel when said final pixel exceeds said predetermined constant value.

17. The method according to claim 16, comprising:

repeating said steps (b) and (c) for pixels preceding said final pixel until a second pixel is encountered that does not exceed said predetermined constant value.

18. The method according to claim 1, wherein said pre-processing step comprises:

identifying and suppressing direct-exposure and subcutaneous regions from said first image data.

19. The method according to claim 18, wherein said identifying and suppressing step comprises:

setting pixels subsequent to an initial pixel consecutively in each row to zero until a first pixel is encountered with a value exceeding a predetermined multiple of the smallest pixel value encountered subsequent to said initial pixel; and

calculating an average of N pixels subsequent to said first pixel in the same row; and

setting pixels subsequent to said first pixel consecutively to zero until a second pixel is encountered with a value exceeding said average of N pixels subsequent to said first pixel.

20. The method according to claim 18, further comprising:

setting pixels prior to a final pixel to zero until a third pixel is encountered with a value exceeding a predetermined multiple of the smallest pixel value encountered since said final pixel; and

calculating an average of N pixels prior to said third pixel in the same row; and

setting pixels prior to said third pixel to zero until a fourth pixel is encountered with a value exceeding said average of N pixels prior to said third pixel.

21. The method according to claim 1, wherein said pre-processing step further comprises:

locating a lung apex position in said first image data.

22. The method according to claim 21, wherein said locating step comprises:

calculating N average rows based on an average of a predetermined number of rows in said second image data;

constructing N horizontal gray-level profiles from said N average rows; and

performing a running mean operation to smooth said N horizontal gray-level profiles and produce smoothed N horizontal gray-level profiles.

23. The method according to claim 22, wherein said locating step comprises:

performing a slope technique to determine a plurality of gray-level maxima and a plurality of gray-level minima in said smoothed N horizontal gray-level profiles; and

searching said smoothed N horizontal gray-level profiles for a first profile representing the superiormost of said smoothed N horizontal gray-level profiles with a lowest gray-level minimum positioned between two gray-level maxima with gray levels exceeding a predetermined multiple of the value of said gray-level minimum.

24. The method according to claim 23, wherein said locating step further comprises:

identifying as a lung apex y-position the row of said first profile; and

identifying as a lung apex x-position a column containing said lowest gray-level minimum of said first profile.

25. The method according to claim 24, wherein said delineating step comprises:

identifying as a first anterior margin point in an initial row in said second image data below said lung apex y-position the first pixel extending posteriorly to the anterior side encountered with a value below a predetermined multiple of a largest pixel value between the anterior side of said initial row and said first anterior margin point; and

identifying as a first posterior margin point in said initial row in said second image data the first pixel extending anteriorly to the posterior side encountered with a value below a predetermined multiple of a largest pixel value between the posterior side of said horizontal gray-level profile and said first posterior margin point.

26. The method according to claim 25, comprising:

repeating said constructing and identifying steps for subsequent rows in said second image data until a collection of said first anterior margin points and a collection of said first posterior margin points is obtained.

27. The method according to claim 26, wherein said delineating step further comprises:

smoothing said collection of first anterior margin points and said collection of first posterior margin points independently through a running mean technique to obtain a collection of second anterior margin points and a collection of second posterior margin points.

28. The method according to claim 27, wherein said smoothing step comprises:

determining for each row respective average locations for the first anterior margin points and the first posterior margin points, based on the average of locations of the first anterior martin points and first posterior margin points in plural adjacent rows including the row for which the respective average locations are to be determined; and

replacing the location of each of said first anterior margin points and said first posterior margin points with the respective average locations to produce said second anterior margin points and said second posterior margin points.

29. The method according to claim 28, comprising:

iteratively repeating said determine step and said replacing step until pluralities of said second anterior margin points and pluralities of said second posterior margin points define continuous segments that exceed a predetermined constant value in length.

30. The method according to claim 29, wherein said delineating step further comprises:

setting to zero in said second image data all pixels that lie anterior to said second anterior margin points and posterior to said second posterior margin points.

31. The method according to claim 1, wherein said delineating step comprises:

applying a Sobel filter to the bottom half of said first image data to create Sobel-filtered image data comprising,

setting the values of pixels in said second image data to a first predetermined constant value when the corresponding pixel in said Sobel-filtered image data exceeds a second predetermined constant value.

32. The method according to claim 1, wherein said performing iterative global gray-level thresholding step comprises:

identifying a range of gray-level threshold values; and

selecting N gray-level threshold values from said range of gray-level threshold values.

33. The method according to claim 32, wherein said identifying step comprises:

constructing a global gray-level histogram from the pixels contained within a large region-of-interest of predetermined dimension;

analyzing the slope of said global gray-level histogram to identify a lung peak comprised of pixels belonging predominantly to the lung and a minimum between said lung peak and a peak comprised of pixels belonging predominantly to the spine, sternum, shoulder, and subdiaphragmatic regions; and

identifying as lower bound of said range of gray-level threshold values the gray level at which said lung peak occurs in said global gray-level histogram;

identifying as an upper bound of said range of gray-level threshold values the gray level at which said minimum occurs in said global gray-level histogram.

34. The method according to claim 32, wherein said performing iterative global gray-level thresholding step further comprises:

creating a first binary image based on said second image data by turning “

on”

pixels in said second image data with values less than the first of said N gray-level threshold values and greater than a predetermined lower limit value;

identifying regions of contiguous “

on”

pixels; and

calculating a center-of-mass in each of said regions of contiguous “

on”

pixels; and

determining which of said regions of contiguous “

on”

pixels exist outside of the lung region;

suppressing regions of contiguous “

on”

pixels in said second image data determined to exist outside the lung regions.

35. The method according to claim 34, wherein said determining step comprises:

generating a horizontal gray-level profile through the center-of-mass of each of said regions of contiguous “

on”

pixels;

identifying maxima and minima in said gray-level profiles using a slope technique;

comparing the gray-level value of said horizontal gray-level profile at a position of the center-of-mass with the gray-level values of said maxima and minima to determine those of said regions of contiguous “

on”

pixels that exist outside the lung regions.

36. The method according to claim 35, wherein said suppressing step comprises:

setting to zero the pixels of said second image data corresponding to those pixels within said regions of contiguous “

on”

pixels in said first binary image determined to exist outside the lung regions;

further comprising;

repeating said creating, identifying, calculating, generating, identifying maxima and minima, and comparing steps for each of said N gray-level threshold values; and

repeating said setting to zero step for all said regions of contiguous “

on”

pixels.

37. The method according to claim 34, 35 or 36, wherein said step of performing iterative global gray-level thresholding is repeated for each of said N gray-level threshold values.

38. The method according to claim 1, wherein said performing iterative global gray-level thresholding step comprises:

outlining said first initial lung segmentation contour around a region of contiguous “

on”

pixels that results after N iterations of global gray-level thresholding.

39. The method according to claim 1, wherein said smoothing step comprises:

applying a rolling ball filter to said first initial lung segmentation contour to produce said second initial lung segmentation contour.

40. The method according to claim 39, wherein said applying step comprises:

constructing a circular filter of a predetermined constant value in radius;

calculating a slope of a line tangent to each point of said first initial lung segmentation contour;

placing a contact point found on the perimeter of the circular filter successively on each point of said first initial lung segmentation contour such that the slope of a line tangent to the circle perimeter of said circular filter at said contact point matches said slope of a line tangent to each point of said first initial lung segmentation contour; and

applying linear interpolation to bridge concavities identified when said ball filter contacts said first initial lung segmentation contour at more than one point.

41. The method according to claim 1, wherein said smoothing step further comprises:

performing a running mean operation that replaces the position of each point of said first initial lung segmentation contour with an average position of a predetermined constant number of adjacent contour points to construct a third initial lung segmentation contour; and

eliminating points in said third initial lung segmentation contour that are redundant in an eight-point connectivity sense to construct said second initial lung segmentation contour.

43. An image processing apparatus configured to perform each of the steps recited in one of claims 1, 2, 14, 17, 20, 22, 31, 26, 29, 33, 36, 40, 4, 8, and 10.

44. A storage medium storing a program for performing each of the steps recited in one of claims 1, 2, 14, 17, 20, 22, 31, 26, 29, 33, 36, 40, 4, 8, and 10.

42. A system for the automated segmentation of the lung region in lateral chest radiographic images, comprising:

means for obtaining image data that represents the thorax of a laterally positioned patient;

means for delineating margins that encompass the portion of image data;

means for performing iterative global gray-level thresholding and adaptive local gray-level thresholding to produce a lung segmentation contour

means for smoothing the lung segmentation contour; and

means for indicating a final contour on a lateral chest image.