COMPUTER AIDED DIAGNOSIS OF LUNG DISEASE

• US 20080082017A1
• Filed: 08/25/2006
• Published: 04/03/2008
• Est. Priority Date: 08/25/2006
• Status: Active Grant

• Current Assignee: Savic Research, LLC
• Original Assignee: Savic Research, LLC
• Inventors: SAVIC, Michael
• Granted Patent: US 7,479,115 B2
• Time in Patent Office: Days
• US Class Current: 600/529
• 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: A61B 5/7267 involving training the classification device
• CPC Class Codes: A61B 7/003 Detecting lung or respiration noise

Abstract

A method and apparatus by which lung deceases are identified uses computer analysis of sound signals that are picked up from various locations on the chest walls of a subject by a modified stethoscope. The modification includes a small microphone in one of the hoses of the stethoscope. Signals from the microphone are input to a computer such as a personal computer or PC for processing. The computer extracts from these signals features which are dominant for particular lung diseases. A classifier classifies these features, determines if the lungs are diseased, and identifies the disease.

Claims

1. A method of identifying a lung disease that produces sounds that are characteristic of the lung disease, comprising subjecting the sounds to autoregressive modeling to recognize and thus identify the lung disease.

2. A method according to claim 1, wherein the use of autoregressive modeling includes identifying autocorrelation coefficients for the sounds to recognize the lung disease.

3. A method according to claim 2, including using at least one of LPC and PARCOR coefficients as the autocorrelation coefficients.

4. A method according to claim 1, including projecting at least one two-dimensional representation of autocorrelation coefficients of the autoregressive modeling for the sounds and identifying the disease by selection at least one pair of coefficients that separate clusters of samples in the representation.

5. A method according to claim 4, wherein the coefficients are at least one of LPC and PARCOR coefficients.

6. A method according to claim 1, including projecting a plurality of two-dimensional representations of autocorrelation coefficients of the autoregressive modeling for the sounds, selected features of the representations that provide relatively good correlation to the lung disease as compared to feature that give relatively poor correlation to the lung disease, and identifying the disease by selection at least one pair of coefficients that separate clusters of samples in the representation for the features that provide relatively good correlation to the lung disease.

7. A method according to claim 1, including projecting at least one two-dimensional representation of autocorrelation coefficients of the autoregressive modeling for the sounds, subjecting the representation to discriminant analysis to select dominant features of the representation and using neural network analysis of the dominant features for identifying the disease.

8. A method according to claim 1, using discriminant analysis to select dominant features of the autoregressive modeling and using neural network analysis of the for dominant features for identifying the disease.

9. A method according to claim 1, including storing in a computer, a data base of dominant features from pre-recorded sounds from lungs with at least one lung disease during a training phase, obtaining sounds from lungs of a subject to be tested, conditioning the sounds to create a signal that can be subjected to analysis in the computer, using the autoregressive modeling on the signal in the computer and using the data base with autoregressively modeled signal to identify the disease.

10. A method according to claim 1, wherein the use of autoregressive modeling includes extracting a plurality of autocorrelation coefficients from the sounds, subjecting at least some of the coefficients to multi-dimensional projection to extract features of the projection, and using the features to identifying the lung disease.

11. An apparatus for identifying a lung disease in a subject, the lung disease producing lung sounds that are characteristic of the lung disease, the apparatus comprising:

a stethoscope for acquiring the lung sounds from the subject;

a transducer operatively connected to the stethoscope for converting the lung sounds to a signal; and

a computer-based analyzer for analyzing the signal to identify the lung disease.

12. An apparatus according to claim 11, wherein the stethoscope includes at least one hose for conveying the sounds, and the transducer comprises a microphone connected to the hose for picking up sounds in the hose.

13. An apparatus according to claim 11, wherein the computer-based analyzer comprises a computer that is programmed for analyzing the signal to identify the lung disease using autoregressive modeling.

14. An apparatus according to claim 11, wherein the computer-based analyzer comprises a computer that is programmed with a program for analyzing the signal to identify the lung disease using autoregressive modeling.

15. An apparatus according to claim 14, wherein the program identifies autocorrelation coefficients of the autoregressive modeling for the signal to identify the lung disease.

16. An apparatus according to claim 15, wherein the program uses at least one of LPC and PARCOR coefficients as the autocorrelation coefficients.

17. An apparatus according to claim 14, wherein the program plots at least one two-dimensional representation of autocorrelation coefficients of the autoregressive modeling for the signal and identifies the disease by selection at least one pair of coefficients that separate clusters of samples in the representation.

18. An apparatus according to claim 17, wherein the coefficients are at least one of LPC and PARCOR coefficients.

19. An apparatus according to claim 14, wherein the program plots a plurality of two-dimensional representations of autocorrelation coefficients of the autoregressive modeling for the sounds, and selects features of the representations that provide relatively good correlation to the lung disease as compared to feature that give relatively poor correlation to the lung disease, and identifies the disease by selection at least one pair of coefficients that separate clusters of samples in the representation for the features that provide relatively good correlation to the lung disease.

20. An apparatus according to claim 14, wherein the program plots at least one two-dimensional representation of autocorrelation coefficients of the autoregressive modeling for the sounds, and subjects the representation to discriminant analysis to select dominant features of the representation, and uses neural network analysis of the dominant features for identifying the disease.

21. An apparatus according to claim 14, wherein the program uses discriminant analysis to select dominant features of the autoregressive modeling and using neural network analysis of the for dominant features for identifying the disease.

22. An apparatus according to claim 11, a data base stored in the computer-based analyzer, of dominant features from pre-recorded sounds from lungs with at least one lung disease during a training phase, means for conditioning the sounds to create the signal that can be subjected to analysis in the analyzer, the analyzer being programed to perform autoregressive modeling on the signal and to use the data base with autoregressively modeled signal to identify the disease.

23. An apparatus according to claim 11, wherein the program extracts a plurality of autocorrelation coefficients from the sounds, subjects at least some of the coefficients to multi-dimensional projection to extract features of the projection, and uses the features to identifying the lung disease.

24. An apparatus for identifying a disease in a subject, the disease producing sounds that are characteristic of the disease, the apparatus comprising:

means for acquiring the sounds; and

a computer-based analyzer for analyzing the sounds to identify the disease using autoregressive modeling.

25. An apparatus according to claim 24, wherein the computer-based analyzer comprises a computer that is programmed with a program for analyzing the sounds to identify the disease using autoregressive modeling, the program using at least one of LPC and PARCOR coefficients as autocorrelation coefficients for the modeling.

26. An apparatus according to claim 24, wherein the program projects at least one two-dimensional representation of autocorrelation coefficients of the autoregressive modeling for the sound and identifies the disease by selection at least one pair of coefficients that separate clusters of samples in the representation.

27. An apparatus according to claim 26, wherein the coefficients are at least one of LPC and PARCOR coefficients.

28. An apparatus according to claim 24, wherein the program plots a plurality of two-dimensional representations of autocorrelation coefficients of the autoregressive modeling for the sounds, and selects features of the representations that provide relatively good correlation to the lung disease as compared to feature that give relatively poor correlation to the lung disease, and identifies the disease by selection at least one pair of coefficients that separate clusters of samples in the representation for the features that provide relatively good correlation to the disease.

29. An apparatus according to claim 24, wherein the program plots at least one two-dimensional representation of autocorrelation coefficients of the autoregressive modeling for the sounds, and subjects the representation to discriminant analysis to select dominant features of the representation, and uses neural network analysis of the dominant features for identifying the disease.

30. An apparatus according to claim 24, wherein the program uses discriminant analysis to select dominant features of the autoregressive modeling and using neural network analysis of the for dominant features for identifying the disease.

31. An apparatus according to claim 24, including a data base stored in the computer-based analyzer, of dominant features from pre-recorded sounds from lungs with at least one lung disease during a training phase, means for conditioning the sounds to create the signal that can be subjected to analysis in the analyzer, the analyzer being programed to perform autoregressive modeling on the signal and to use the data base with autoregressively modeled signal to identify the disease.

32. An apparatus according to claim 24, wherein the program extracts a plurality of autocorrelation coefficients from the sounds, subjects at least some of the coefficients to multi-dimensional projection to extract features of the projection, and uses the features to identifying the disease.