Method and system for automated classification of distinction between normal lungs and abnormal lungs with interstitial disease in digital chest radiographs

• US 4,839,807 A
• Filed: 11/27/1987
• Issued: 06/13/1989
• Est. Priority Date: 08/03/1987
• Status: Expired due to Term

• Current Assignee: University of Chicago The (University of Chicago)
• Original Assignee: University of Chicago
• Inventors: Katsuragawa, Shigehiko, Doi, Kunio
• Primary Examiner(s): Smith, Jerry
• Assistant Examiner(s): Hayes, Gail O.
• Application Number:
  US07/126,847
• Time in Patent Office: 564 Days
• Field of Search: 364/414, 378/401, 382/4, 382/6
• US Class Current: 382/128
• CPC Class Codes: A61B 5/08 Detecting, measuring or recording devices for evaluating the respiratory organs A61B5/0205 takes precedence
• CPC Class Codes: A61B 5/7264 Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems neural networks per se G06N3/00; expert systems per se G06N5/00
• CPC Class Codes: G06T 2207/30061 Lung
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/41 based on statistical description of texture
• CPC Class Codes: G16H 15/00 ICT specially adapted for medical reports, e.g. generation or transmission thereof
• 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: G16H 70/60 relating to pathologies
• CPC Class Codes: Y10S 378/901 Computer tomography program or processor

Abstract

A method and system for automated classification of distinction between normal lungs and abnormal lungs with interstitial disease, based on the analysis of predetermined physical texture measures and also on a data base for normal lungs of these texture measures. The texture measures selected are the RMS variation, R, and the first moment of power spectrum, M, for lung texture. These two texture measures are normalized by using the data base for normal lungs. A single texture index is determined from the two normalized texture measures by taking into account the distribution (or the data base) of texture measures obtained from abnormal lungs, in order to facilitate the automated classification of normal and abnormal lungs. A threshold texture index is then chosen for initial selection of "abnormal" regions of interest (ROIs), which contain a large texture index above the threshold level. The selected abnormal ROIs are then subjected to three independent tests for a (1) definitely abnormal singular pattern, (2) localized abnormal pattern for two or more abnormal clustered ROIs, and (3) diffuse abnormal pattern for more than four abnormal ROIs distributed through the lung. A chest image containing any one of these abnormal patterns is classified as showing an abnormal lung with interstitial disease.

Claims

1. A method for automated classification of distinction between normal and abnormal lungs with interstitial disease in digital chest radiographs, comprising:

sampling plural regions of interest (ROIs) of a subject digital radiograph;

producing digital data indicative of the texture of each of said ROIs;

determining from the digital data of each selected ROI texture measures indicative of the lung texture of the respective ROIs;

normalizing the texture measures determined in said determining step in relation to predetermined characteristics derived from a data base of normal lungs; and

processing normalized texture measures obtained in said normalizing step to determine based on predetermined criteria whether or not a lung image of the subject digital radiograph is normal or abnormal.

2. The method according to claim 1, wherein said determining step comprises:

removing non-uniform background trend in the digital data of each ROI in order to isolate underlying fine texture;

determining the power spectrum of the underlying fine texture;

bandpass filtering the determined power spectrum;

determining the root-mean-square (RMS) variation (R) and the first moment of the power spectrum (M) of the bandpass filtered power spectrum; and

using R and M as said predetermined texture measures.

3. The method according to claim 2, wherein said normalizing step comprises:

determining normalized values of R_N and M_N based on the following relationships, ##EQU4## where R^- and M^- are respectively the average RMS variation and the average first moment of the power spectrum for normal lungs as derived from said data base, and σ

_R and σ

_M are respectively the standard deviations of the RMS variation and the first moment of the power spectrum for normal lungs, respectively.

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

calculating a single texture index (T) based on the normalized values of R_N and M_N.

5. The method according to claim 4, wherein T is defined in (M_N,R_N) coordinates as follows:

T=R_N, for a first quadrant where M_N >

0 and R_N >

0,

T=√

M_N² +R_N², for a second quadrant where M_N <

0 and R_N >

0,

T=-M_N, for a third quadrant where M_N <

0 and R_N <

0, and

T=-{Min(M_N, |R_N |)} for a fourth quadrant where M_N >

0 and R_N <

0.

6. The method according to claim 4, wherein said processing step comprises:

comparing the texture index T of each selected ROI against a first threshold value; and

determining whether or not each ROI is potentially abnormal if the respective texture index exceeds said first threshold value in said comparing step.

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

comparing the texture index T of those ROIs determined to be potentially abnormal against a second threshold value; and

determining the existence of an abnormal ROI pattern if any texture index T exceeds said second threshold value.

8. The method according to claim 7, wherein said processing step comprising:

determining whether those ROIs determined to be potentially abnormal are clustered within a predetermined distance of each other; and

determining the existence of an abnormal ROI pattern if it is determined that the potentially abnormal ROIs are clustered within said predetermined distance of each other.

9. The method according to claim 8, wherein said processing step comprises:

counting the number of said potentially abnormal ROIs; and

determining the existence of an abnormal ROI pattern if the number of potentially abnormal ROIs exceeds a third predetermined threshold value.

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

determining whether any abnormal ROIs are reticular, nodular, or honeycomb and reticulo-nodular based on the respective quadrant of (M_N, R_N) coordinates and the location within a quadrant of said (M_N, R_N) coordinates of the ROIs.

11. The method according to claim 10, comprising:

displaying the image formed on said digital radiograph; and

identifying on the displayed image those ROIs corresponding to any abnormal ROI pattern.

12. The method according to claim 11, comprising:

displaying respective abnormality symbols identifying particular abnormal ROIs of abnormal ROI patterns as being reticular, nodular, or honeycomb and reticulo-nodular on the displayed image at the location of the respective abnormal ROI in the displayed image.

13. The method according to claim 12, comprising:

varying the size of each abnormality symbol in relation to the magnitude of the texture index T of the respective ROI.

14. The method according to claim 13, comprising:

deriving a receiver operating characteristic (ROC) curve as a function of said first threshold value in terms of the variation of a true positive fraction versus a false-positive fraction of ROIs determined to be abnormal; and

displaying a true-positive fraction and a false-positive fraction in addition to said image and said abnormality symbols.

15. The method according to claim 14, further comprising:

interactively varying said first threshold value and displaying said image, said abnormality symbols and said true-positive and false-positive fractions as determined for a varied first threshold value.

16. The method according to claim 15, further comprising:

interactively varying said first threshold value and displaying said image, said abnormality symbols and said true-positive and false-positive fractions as determined for a varied first threshold value.

17. The method according to claim 13, comprising:

varying the size of each abnormality symbol in relation to the magnitude of the texture index T of the respective ROI.

18. The method according to claim 17, comprising:

deriving a receiver operating characteristic (ROC) curve as a function of said first threshold value in terms of the variation of a true positive fraction versus a false-positive fraction of ROIs determined to be abnormal; and

displaying a true-positive fraction and a false-positive fraction in addition to said image and said abnormality symbols.

19. The method according to claim 8, wherein said processing step comprises:

determining whether any abnormal ROIs are reticular, nodular, or honeycomb and reticulo-nodular based on the respective quadrant of (M_N, R_N) coordinates and the location within a quadrant of said (M_N, R_N) coordinates of the ROIs.

20. The method according to claim 19, comprising:

displaying the image formed on said digital radiograph; and

identifying on the displayed image those ROIs corresponding to any abnormal ROI pattern.

21. The method according to claim 20, comprising:

displaying respective abnormality symbols identifying particular abnormal ROIs of abnormal ROI patterns as being reticular, nodular, or honeycomb and reticulo-nodular on the displayed image at the location of the respective abnormal ROI in the displayed image.

22. The method according to claim 7, wherein said processing step comprises:

determining whether any abnormal ROIs are reticular, nodular, or honeycomb and reticulo-nodular based on the respective quadrant of (M_N, R_N) coordinates and the location within a quadrant of said (M_N, R_N) coordinates of the ROIs.

23. The method according to claim 6, wherein said processing step comprising:

determining whether those ROIs determined to be potentially abnormal are clustered within a predetermined distance of each other; and

determining the existence of an abnormal ROI pattern if it is determined that the potentially abnormal ROIs are clustered within said predetermined distance of each other.

24. The method according to claim 6, wherein said processing step comprises:

counting the number of said abnormal ROIs; and

determining the existence of an abnormal ROI pattern if the number of potentially abnormal ROIs exceeds a third predetermined threshold value.

25. A system for automated classification of distinction between normal and abnormal lungs with interstitial disease in digital chest radiographs, comprising:

means for sampling plural regions of interest (ROIs) of a subject digital radiograph for evaluation;

means for producing digital data indicative of the texture of each of said ROIs;

means for determining from the digital data of each selected ROI texture measures indicative of the lung texture of the respective ROIs;

means for normalizing the texture measures determined by said determining step in relation to predetermined characteristics derived from a data base of normal lungs; and

means for processing normalized texture measures obtained by said normalizing means to determine based on predetermined criteria whether or not a lung image of the subject digital radiograph is normal or abnormal.

26. The system according to claim 25, wherein said determining means comprises:

means for removing non-uniform background trend in the digital data of each ROI in order to isolate underlying fine texture;

means for determining the power spectrum of the underlying fine texture;

means for bandpass filtering the determined power spectrum;

means for determining the root-mean-square (RMS) variation (R) and the first moment of the power spectrum (M) of the bandpass filtered power spectrum;

wherein R and M are used as said predetermined texture measures.

27. The system according to claim 26, wherein said normalizing means comprises:

means for determining normalized values of R_N and M_N based on the following relationships, ##EQU5## where R^- and M^- are respectively the average RMS variation and the average first moment of the power spectrum for normal lungs as derived from said data base, and σ

_R and σ

_M are respectively the standard deviations of the RMS variation and the first moment of the power spectrum for normal lungs, respectively.

28. The system according to claim 27, wherein said processing means comprises:

means for calculating a single texture index (T) based on the normalized values of R_N and M_N ;

wherein R is defined in (M_N, R_N) coordinates as follows;

T=R_N, for a first quadrant where M_N >

0 and R_N >

0,

T=√

M_N² +R_N² for a second quadrant where M_N <

0 and R_N >

0,

T=-M_N, for a third quadrant where M_N <

0 and R_N <

0, and

T=-{Min(M_N, |R_N |)} for a fourth quadrant where M_N >

0 and R_N <

0.

29. The system according to claim 28, wherein said processing means comprises:

means for comparing the texture index T of each selected ROI against a first threshold value;

wherein it is determined whether or not each ROI is potentially abnormal if the respective texture index exceeds said first threshold value.

30. The system according to claim 29, wherein said processing means comprises:

means for comparing the texture index T of those ROIs determined to be potentially abnormal against a second threshold value; and

wherein the existence of an abnormal ROI pattern is determined if any texture index T exceeds said second threshold value.

31. The system according to claim 30, wherein said processing means comprising:

means for determining whether those ROIs determined to be potentially abnormal are clustered within a predetermined distance of each other;

wherein the existence of an abnormal ROI pattern is determined if it is determined that the potentially abnormal ROIs are clustered within said predetermined distance of each other.

32. The system according to claim 31, wherein said processing means comprises:

means for counting the number of said abnormal ROIs; and

means for determining the existence of an abnormal ROI pattern if the number of potentially abnormal ROIs exceeds a third predetermined threshold value.

33. The system according to claim 32, wherein said processing means comprises:

means for determining whether an ROI of any abnormal pattern of ROIs is reticular, nodular, or honeycomb and reticulo-nodular based on the respective quadrant of (M_N, R_N) coordinates and the location within a quadrant of said (M_N, R_N) coordinates of the ROI.

34. The system according to claim 33, comprising:

means for displaying the image formed on said digital radiograph; and

means for identifying on the displayed image those ROIs corresponding to any abnormal ROI pattern.

35. The system according to claim 34, comprising:

means for generating respective abnormality symbols identifying particular abnormal ROIs of abnormal ROI patterns as being reticular, nodular, or honeycomb and reticulo-nodular;

wherein said abnormality symbols are displayed on the displayed image at the location of the respective abnormal ROI in the displayed image.

36. The system according to claim 33, comprising:

means for varying the size of each abnormality symbol in relation to the magnitude of the texture index T of the respective ROI.

37. The system according to claim 36, comprising:

means for storing a receiver operating characteristic (ROC) curve as a function of said first threshold value in terms of the variation of a true positive fraction versus a false-positive fraction of ROIs determined to be abnormal; and

means for displaying a true-positive fraction and a false-positive fraction in addition to said image and said abnormality symbols.

38. The system according to claim 37, further comprising:

means for interactively varying said first threshold value so that said abnormality symbols and said true-positive and false-positive fractions vary a function of the varied first threshold value and are displayed in relation to a varied first threshold value.

39. The system according to claim 29, wherein said processing means comprising:

means for determining whether those ROIs determined to be potentially abnormal are clustered within a predetermined distance of each other;

wherein the existence of an abnormal ROI pattern is determined if it is determined that the potentially abnormal ROIs are clustered within said predetermined distance of each other.

40. The system according to claim 29, wherein said processing means comprises:

means for counting the number of said abnormal ROIs; and

means for determining the existence of an abnormal ROI pattern if the number of potentially abnormal ROIs exceeds a third predetermined threshold value.

41. The system according to claims 30, 39 or 40, wherein said processing means comprises:

means for determining whether an ROI of any abnormal pattern of ROIs is reticular, nodular, or honeycomb and reticulo-nodular based on the respective quadrant of (M_N, R_N) coordinates and the location within a quadrant of said (MN, RN) coordinates of the ROI.

42. The system according to claim 41, comprising:

means for displaying the image formed on said digital radiograph; and

means for identifying on the displayed image those ROIs corresponding to any abnormal ROI pattern.

43. The system according to claim 42, comprising:

means for generating respective abnormality symbols identifying particular abnormal ROIs of abnormal ROI patterns as being reticular, nodular, or honeycomb and reticulo-nodular;

wherein said abnormality symbols are displayed on the displayed image at the location of the respective abnormal ROI in the displayed image.

44. The system according to claim 43, comprising:

means for varying the size of each abnormality symbol in relation to the magnitude of the texture index T of the respective ROI.

45. The system according to claim 44, comprising:

means for storing a receiver operating characteristic (ROC) curve as a function of said first threshold value in terms of the variation of a true positive fraction versus a false-positive fraction of ROIs determined to be abnormal; and

means for displaying a true-positive fraction and a false-positive fraction in addition to said image and said abnormality symbols.

46. The system according to claim 45, further comprising:

means for interactively varying said first threshold value so that said abnormality symbols and said true-positive and false-positive fractions vary as a function of the varied first threshold value and are displayed in relation to a varied first threshold value.