Method and system for automated selection of regions of interest and detection of septal lines in digital chest radiographs

• US 5,343,390 A
• Filed: 02/28/1992
• Issued: 08/30/1994
• Est. Priority Date: 02/28/1992
• Status: Expired due to Term

• Current Assignee: ARCH Development Corporation (University of Chicago)
• Original Assignee: ARCH Development Corporation (University of Chicago)
• Inventors: Chen, Xuan, Doi, Kunio, Katsuragawa, Shigehiko
• Primary Examiner(s): Envall, Jr., Roy N.
• Assistant Examiner(s): Tkacs, Stephen R.
• Application Number:
  US07/843,721
• Time in Patent Office: 914 Days
• Field of Search: 358/111, 364/413.13, 364/413.16, 364/413.22, 364/413.23, 378/99, 378/901, 382/6, 382/22, 382/52, 382/54
• US Class Current: 382/132
• CPC Class Codes: G06T 2207/10116 X-ray image
• CPC Class Codes: G06T 2207/20132 Image cropping
• CPC Class Codes: G06T 2207/30061 Lung
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G16H 30/20 for handling medical images, e.g. DICOM, HL7 or PACS
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: Y10S 378/901 Computer tomography program or processor

Abstract

An automated method and system for discriminating between normal lungs and abnormal lungs having interstitial disease and/or septal lines, wherein a large number of adjacent regions of interest (ROIs) are selected, corresponding to an area on a digital image of a patient'"'"'s lungs. The ROIs each contain a number of square or rectangular pixel arrays and are selected to sequentially fill in the total selected area of the lungs to be analyzed. A background trend is removed from each individual ROI and the ROIs are then analyzed to determine those exhibiting sharp edges, i.e., high edge gradients. A percentage of these sharp-edged ROIs are removed from the original sample based on the edge gradient analysis, a majority of which correspond to rib-edge containing ROIs. After removal of the sharp-edged ROIs, texture measurements are taken on the remaining sample in order to compare such data with predetermined data for normal and abnormal lungs. Thus, a computerized scheme for quantitative analysis of interstitial lung diseases and/or septal lines appearing in digitized chest radiographs can be implemented in practical clinical situations.

Claims

1. A computer implemented method for automated detection of abnormalities in digital chest radiographs comprising:

identifying peripheral lung areas of a digital chest radiograph, stored in an image memory in a computer, to be analyzed;

preselecting, using said computer, a sample of numerous contiguous regions of interest (ROIs) included in said peripheral lung areas;

performing a background trend correction, using said computer, on each of the ROIs of said preselected sample so as to produce corrected image data;

performing an edge gradient analysis, using said computer, on each of the ROIs of said preselected sample in order to distinguish sharp-edged ROIs from ROIs with sharp edges;

removing a portion of said sharp-edged ROIs from said preselected sample; and

performing texture measurements on the remaining ROIs of said preselected sample which have not been removed in said removing step.

2. The method according to claim 1, wherein said step of preselecting a sample of numerous contiguous ROIs comprises:

selecting numerous square ROIs, each having a matrix size which is small enough to fit between a pair of adjacent ribs, wherein said ROIs are selected so as to be located within the peripheral lung areas in order to form a grid of adjacent square ROIs covering nearly the entire peripheral lung areas.

3. The method according to claim 1, wherein said step of preselecting a sample of numerous contiguous ROIs comprises:

initially selecting approximately 400 to 600 ROIs so as to form a grid of adjacent square ROIs covering nearly the entire peripheral lung areas.

4. The method according to claim 1, wherein said step of identifying peripheral lung areas comprises:

selecting an upper boundary of a peripheral lung area at a distance below the lung apex, said distance equal to approximately 20% of the lung height;

selecting a lower boundary of said peripheral lung area to be the top of the diaphragm;

selecting an outer boundary of said peripheral lung at a location approximately 96% of the distance from a midline location of said chest radiograph to an edge of the rib cage; and

selecting an inner boundary which changes linearly from 50% of the distance from said midline of the chest radiograph image to the rib cage edge up to 80% at the lower boundary.

5. The method according to claim 1, wherein said step of performing an edge gradient analysis comprises:

detecting edge gradients using a gradient-weighted edge orientation histogram for detecting step edges or line edges in each of the ROIs of said preselected sample.

6. The method according to claim 5, comprising:

calculating a standard deviation of said gradient-weighted edge orientation histogram for use as a measurement in determining the sharp-edged ROIs to be removed.

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

eliminating those ROIs exhibiting high edge gradients having standard deviations in a selected upper percentage of standard deviations calculated for edge gradients of each ROI from said preselected sample.

8. The method according to claim 6, wherein said step of removing comprises:

eliminating said sharp-edged ROIs with edge gradient standard deviation values above a predetermined threshold value.

9. The method according to claim 1, wherein said step for performing texture measurements comprises:

determining the RMS variation, R, and the first moment of the power spectrum, M, where R represents the magnitude of the texture of an ROI and M represents the coarseness or fineness of the texture.

10. The method according to claim 1, comprising:

manually selecting individual ROIs from a normal disease-free chest radiograph, wherein said manually selected ROIs are selected from three regions including rib edges, areas over ribs and intercostal areas, in order to obtain predetermined texture measurements and edge gradient analyses of said ROIs from said three regions in order to provide information about ROIs with known locations.

11. The method according to claim 1, comprising:

displaying said texture measurements superimposed on a chest image utilizing various sizes and shapes of markers representing the type and severity of lung textures of said ROIs.

12. A computer system for automated detection of abnormalities in digital chest radiographs, comprising:

means for identifying peripheral lung areas of a digital chest radiograph to be analyzed;

means for preselecting a sample of numerous contiguous regions of interest (ROIs) included in said peripheral lung areas;

means for performing a background trend correction on each of the ROIs of said preselected sample so as to produce corrected image data;

means for performing an edge gradient analysis on each of the ROIs of said preselected sample in order to distinguish sharp-edged ROIs from ROIs without sharp edges;

means for removing a portion of said sharp-edged ROIs from said preselected sample; and

means for performing texture measurements on the remaining ROIs of said preselected sample which have not been removed in said removing step.

13. The system according to claim 12, wherein said means for preselecting a sample of numerous contiguous ROIs comprises:

means for selecting numerous square ROIs having a matrix size small enough to fit between a pair or adjacent ribs, wherein said ROIs are selected so as to be located within the peripheral lung areas in order to form a grid of adjacent square ROIs covering nearly the entire peripheral lung areas.

14. The system according to claim 12, wherein said means for preselecting a sample of numerous contiguous ROIs comprises:

means for initially selecting approximately 400 to 600 ROIs so as to form a grid of adjacent square ROIs covering nearly the entire peripheral lung areas.

15. The system according to claim 12, wherein said means for identifying peripheral lung areas comprises:

means for selecting an upper boundary of a peripheral lung area at a distance below the lung apex, said distance equal to approximately 20% of the lung height;

means for selecting a lower boundary of said peripheral lung area to be the top of the diaphragm;

means for selecting an outer boundary of said peripheral lung area at a location approximately 96% of the distance from a midline of said chest radiograph to an edge of the rib cage; and

means for selecting an inner boundary which changes linearly from 50% of the distance from the midline of said chest radiograph image to the rib cage edge up to 80% at the lower boundary.

16. The system according to claim 12, wherein said means for performing an edge gradient analysis further comprises:

means for detecting edge gradients using a gradient-weighted edge orientation histogram for detecting step edges or line edges in each of the ROIs of said preselected sample.

17. The system according to claim 16, further comprising:

means for calculating a standard deviation of said gradient-weighted edge orientation histogram for use as a measurement in determining the sharp-edged ROIs to be removed.

18. The system according to claim 17, wherein said means for removing comprises:

means for eliminating those ROIs which exhibit high edge gradients having standard deviations in a selected upper percentage of all calculated standard deviations.

19. The system according to claim 17, wherein said means for removing comprises:

means for eliminating said sharp-edged ROIs with standard deviation values above a predetermined threshold value.

20. The system according to claim 12, wherein said means for performing texture measurements comprises:

means for determining the RMS variation, R, and the first moment of the power spectrum, M, where R represents the magnitude of the texture of an ROI and M represents the coarseness or fineness of the texture.

21. The system according to claim 12, comprising:

means for manually selecting individual ROIs from a normal disease-free chest radiograph, wherein said means for manually selecting ROIs selects said ROIs from three regions including rib edge regions, areas over ribs and intercostal areas, in order to obtain predetermined texture measurements and edge gradient analyses of said ROIs from said three regions to provide a database of information about ROIs with known locations.

22. The system according to claim 12, further comprising:

means for displaying said texture measurements superimposed on a chest image utilizing various sizes and shapes of markers representing the type and severity of lung textures of said ROIs.

23. A computer implemented method for automated detection of septal lines in a digital chest radiograph, stored in an image memory in a computer, comprising:

identifying a lower lung area along rib cage edges in said digital chest radiograph;

preselecting, using said computer, a sample of numerous contiguous regions of interest (ROIs) within said lower lung area;

performing contrast enhancement, in said computer, using an unsharp masking technique with a rectangular mask;

removing a background trend from each individual ROI;

calculating accumulated edge gradients for each of said individual ROIs;

generating an edge-gradient orientation histogram, in said computer, for each of said individual ROIs based on said calculated accumulated edge gradients;

comparing, using said computer, each generated edge-gradient orientation histogram with predetermined values stored in a database to obtain a comparison result for each of said individual ROIs;

determining if septal lines are present in each of said individual ROIs based on said comparison result for each of said individual ROIs; and

classifying, using said computer, each of said individual ROIs as normal or abnormal having septal lines.

24. The method according to claim 23, further comprising:

displaying of said abnormal ROIs on a CRT monitor screen.

25. The method according to claim 23, wherein said step of performing an edge gradient analysis comprises:

determining accumulated edge gradients at 90° and

270° and

wherein said edge gradients are determined by use of a Sobel operation.

26. The method according to claim 23, wherein said step of identifying said lower lung area includes determining locations of rib cage and diaphragm boundaries.

27. The method according to claim 26, wherein said locations of rib cage boundaries are determined from minimum values of the second derivatives of horizontal signatures.

28. The method according to claim 26, wherein said locations of said diaphragm boundary are determined from maximum values of the first derivatives of vertical signatures which are then smoothed by a curve-fitting technique with a polynomial function.

29. The method according to claim 23, wherein said sample of ROIs include a plurality of rectangular ROIs comprising 31×

9 pixels which are determined continuously in said lower lung area so as to form a grid of adjacent rectangular ROIs covering nearly the entire lower lung area.

30. The method according to claim 23, wherein said step of classifying said ROIs comprises:

classifying said ROIs as either normal or abnormal using a linear discriminant function which is determined by multivariate analysis of a database of known samples containing normal and abnormal ROIs.

31. The method according to claim 23, wherein said step of removing said background trend comprises using a surface-fitting technique with a 4th-order polynomial function.

32. The method according to claim 23, further comprising plotting ROC curves which indicate the relationship between a true positive fraction corresponding to correct detections of septal lines in abnormal ROIs, and false positive fraction corresponding to incorrect detection of septal lines in normal ROIs as a function of a threshold level.

33. A computer system for automated detection of septal lines in digital chest radiographs, comprising:

means for identifying a lower lung area along rib cage edges;

means for preselecting a sample of numerous contiguous regions of interest (ROIs) within said lower lung area;

means for performing contrast enhancement using an unsharp masking technique with a rectangular mask;

means for removing a background trend from each individual ROI;

means for calculating accumulated edge gradients for each of said individual ROIs;

means for generating an edge-gradient orientation histogram for each individual ROI based on said calculated accumulated edge gradients;

means for comparing each generated edge-gradient orientation histogram with predetermined values stored in a database to obtain a comparison result for each individual ROI;

means for determining if septal lines are present in each of said individual ROIs based on said comparison result for each of said individual ROIs; and

means for classifying each of said individual ROIs as normal or as abnormal having septal lines.

34. The system according to claim 33, further comprising:

CRT monitor means for displaying abnormal ROIs.

35. The system according to claim 33, wherein said means for performing an edge gradient analysis comprises:

means for determining accumulated edge gradients at 90° and

270° and

wherein said edge gradients are determined by use of a Sobel operation.

36. The system according to claim 33, wherein said means for identifying said lower lung areas includes means for determining locations of rib cage and diaphragm boundaries.

37. The system according to claim 36, further comprising means for determining said locations of rib cage boundaries using minimum values of the second derivatives of horizontal signatures.

38. The system according to claim 36, wherein said locations of said diaphragm boundaries are determined from maximum values of the first derivatives of vertical signatures which are then smoothed by a curve-fitting technique with a polynomial function.

39. The system according to claim 33, wherein said sample of ROIs includes a plurality of rectangular ROIs comprising 31×

9 pixels which are determined continuously in said lower lung area so as to form a grid of adjacent rectangular ROIs covering nearly the entire lower lung area.

40. The system according to claim 33, wherein said means for classifying said ROIs comprises:

means for classifying said ROIs as either normal or abnormal using a linear discriminant function which is determined by multivariate analysis of a database of known samples containing normal and abnormal ROIs.

41. The system according to claim 33, wherein said means for removing said background trend comprises a means for performing a surface-fitting technique with a 4th-order polynomial function.

42. The system according to claim 33, further comprising a means for plotting ROC curves which indicate the relationship between a true positive fraction corresponding to correct detection of septal lines in abnormal ROIs and false positive fraction corresponding to incorrect detection of septal lines in normal ROIs as a function of a threshold level.