Computerized detection of lung nodules using energy-subtracted soft-tissue and standard chest images

• US 6,240,201 B1
• Filed: 07/24/1998
• Issued: 05/29/2001
• Est. Priority Date: 07/24/1998
• Status: Expired due to Fees

• Current Assignee: ARCH Development Corporation (University of Chicago)
• Original Assignee: ARCH Development Corporation (University of Chicago)
• Inventors: Xu, Xin-Wei, Doi, Kunio, MacMahon, Heber
• Primary Examiner(s): Bella, Matthew C.
• Assistant Examiner(s): CHOOBIN, BARRY
• Application Number:
  US09/121,719
• Time in Patent Office: 1,040 Days
• Field of Search: 382/130, 382/132, 382/154, 382/128, 382/131, 378/8, 378/37, 128/925, 324/308, 600/364, 600/501, 600/526
• US Class Current: 382/130
• CPC Class Codes: G06T 7/0012 Biomedical image inspection

Abstract

A method, system and computer readable medium configured for computerized detection of lung abnormalities, including obtaining a standard digital chest image and a soft-tissue digital chest image; generating a first difference image from the standard digital chest image and a second difference image from the soft-tissue digital chest image; identifying candidate abnormalities in the first and second difference images; extracting from the standard digital chest image and the first difference image predetermined first features of each of the candidate abnormalities identified in the first difference image; extracting from the soft-tissue digital chest image and the second difference images predetermined second features of each of the candidate abnormalities identified in the second difference image; analyzing the extracted first features and the extracted second features to identify and eliminate false positive candidate abnormalities respectively corresponding thereto; applying extracted features from remaining candidate abnormalities derived respectively from the first and second difference images and remaining after the elimination of the false positive candidate abnormalities to respective artificial neural networks to eliminate further false positive candidate abnormalities; performing a logical OR operation of the candidate abnormalities derived respectively from the first and second difference images and remaining after the elimination of the false positive candidate abnormalities; and outputting a signal indicative of a result of performing the logical OR operation. The logical OR combination, of locations of the candidate abnormalities detected in the first difference image and the second difference image, yields an improved detection sensitivity (over 90%) and only slightly increased false positives rate (3.2 false positives per chest image).

Claims

1. A method for computerized detection of lung abnormalities, comprising:

obtaining first and second digital chest images, said first digital chest image comprising a standard digital chest image and said second digital chest image comprising a soft-tissue digital chest image derived from a chest x-ray image in which bony structures are removed by subtraction of a first weighted low energy x-ra exposed image from a second weighted high energy x-ray exposed image;

generating a first difference image from the standard digital chest image by a step of signal-to-noise ratio (SNR) suppressing filtering of said standard digital chest image to produce a SNR-suppressed standard image, and a step of SNR enhancing filtering of said standard digital chest image to produce a SNR-enhanced standard image, and a step of producing a difference image between said SNR-suppressed standard image and said SNR-enhanced standard image;

generating a second difference image from the soft-tissue digital chest image by a step of signal-to-noise ratio (SNR) suppressing filtering of said soft-tissue digital chest image to produce a SNR-suppressed soft-tissue digital chest image, and a step of SNR enhancing filtering of said soft-tissue digital chest image to produce a SNR-enhanced soft-tissue digital chest image, and a step of producing a difference image between said SNR-suppressed soft-tissue digital chest image and said SNR-enhanced soft-tissue digital chest image;

identifying candidate abnormalities in the first and second difference images;

extracting from the standard digital chest image and the first difference image predetermined first features of each of the candidate abnormalities identified in the first difference image;

extracting from the soft-tissue digital chest image and the second difference images predetermined second features of each of the candidate abnormalities identified in the second difference image;

analyzing the extracted first features and the extracted second features to identify and eliminate false positive candidate abnormalities respectively corresponding thereto;

performing a logical OR operation of the candidate abnormalities derived respectively from the first and second difference images and remaining after the elimination of the false positive candidate abnormalities; and

outputting a signal indicative of a result of performing the logical OR operation.

2. The method of claim 1, wherein the step of outputting comprises:

displaying one of a standard chest image and a soft-tissue chest image corresponding to the respective standard digital chest image and soft-tissue digital chest image, and indicating thereon a location of the candidate abnormalities.

3. The method of claim 1, wherein the step of analyzing the candidate abnormalities comprises:

using adaptive rule-based analysis on plural of the extracted features.

4. The method of claim 1, wherein the step of analyzing the candidate abnormalities comprises:

using adaptive rule-based analysis specific to features extracted from the standard digital chest image and the first difference image to identify and eliminate false positive candidate abnormalities identified in said first difference image; and

using adaptive rule-based analysis specific to features extracted from the soft-tissue digital chest image and the second difference image to identify and eliminate false positive candidate abnormalities identified in said second difference image.

5. The method of claim 1, wherein the step of analyzing the candidate abnormalities comprises performing at least one of the following steps:

applying, for each candidate abnormality derived from the standard digital chest image and the first difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network; and

applying, for each candidate abnormality derived from the soft-tissue digital chest image and the second difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network.

6. The method of claim 3, wherein the step of analyzing the candidate abnormalities comprises performing at least one of the following steps:

applying, for each candidate abnormality derived from the standard digital chest image and the first difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network; and

applying, for each candidate abnormality derived from the soft-tissue digital chest image and the second difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network.

7. The method of claim 4, wherein the step of analyzing the candidate abnormalities comprises performing at least one of the following steps:

applying, for each candidate abnormality derived from the standard digital chest image and the first difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network; and

applying, for each candidate abnormality derived from the soft-tissue digital chest image and the second difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network.

8. A computer readable medium storing computer instructions for computerized detection of lung abnormalities in a standard digital chest image and a soft-tissue digital chest image derived from a chest x-ray image in which bony structures are removed by subtraction of a first weighted low energy x-ray exposed image from a second weighted high energy x-ray exposed image, by performing the steps of:

obtaining first and second digital chest images, said first digital chest image comprising a standard digital chest image and said second digital chest image comprising a soft-tissue digital chest image;

generating a first difference image from the standard digital chest image by a step of signal-to-noise ratio (SNR) suppressing filtering of said standard digital chest image to produce a SNR-suppressed standard image, and a step of SNR enhancing filtering of said standard digital chest image to produce a SNR-enhanced standard image, and a step of producing a difference image between said SNR-suppressed standard image and said SNR-enhanced standard image;

generating a second difference image from the soft-tissue digital chest image by a step of signal-to-noise ratio (SNR) suppressing filtering of said soft-tissue digital chest image to produce a SNR-suppressed soft-tissue digital chest image, and a step of SNR enhancing filtering of said soft-tissue digital chest image to produce a SNR-enhanced soft-tissue digital chest image, and a step of producing a difference image between said SNR-suppressed soft-tissue digital chest image and said SNR-enhanced soft-tissue digital chest image;

identifying candidate abnormalities in the first and second difference images;

extracting from the standard digital chest image and the first difference image predetermined first features of each of the candidate abnormalities identified in the first difference image;

extracting from the soft-tissue digital chest image and the second difference images predetermined second features of each of the candidate abnormalities identified in the second difference image;

analyzing the extracted first features and the extracted second features to identify and eliminate false positive candidate abnormalities respectively corresponding thereto;

performing a logical OR operation of the candidate abnormalities derived respectively from the first and second difference images and remaining after the elimination of the false positive candidate abnormalities; and

outputting a signal indicative of a result of performing the logical OR operation.

9. The computer readable medium of claim 8, further storing computer instructions, for performing the step of outputting, comprising:

displaying one of a standard chest image and a soft-tissue chest image corresponding to the respective standard digital chest image and soft-tissue digital chest image, and indicating thereon a location of the candidate abnormalities derived from the logical OR operation.

10. The computer readable medium of claim 8, wherein the stored computer instructions for performing the step of analyzing comprise:

using adaptive rule-based analysis on plural of the extracted features.

11. The computer readable medium of claim 8, wherein the stored computer instructions for performing the step of analyzing comprise:

using adaptive rule-based analysis specific to features extracted from the standard digital chest image and the first difference image to identify and eliminate false positive candidate abnormalities identified in said first difference image; and

using adaptive rule-based analysis specific to features extracted from the soft-tissue digital chest image and the second difference image to identify and eliminate false positive candidate abnormalities identified in said second difference image.

12. The computer readable medium of claim 8, wherein the stored computer instructions for performing the step of analyzing comprise instructions for performing at least one of the following steps:

applying, for each candidate abnormality derived from the standard digital chest image and the first difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network; and

applying, for each candidate abnormality derived from the soft-tissue digital chest image and the second difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network.

13. The computer readable medium of claim 10, wherein the stored computer instructions for performing the step of analyzing comprise instructions for performing at least one of the following steps:

applying, for each candidate abnormality derived from the standard digital chest image and the first difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network; and

applying, for each candidate abnormality derived from the soft-tissue digital chest image and the second difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network.

14. The computer readable medium of claim 11, wherein the stored computer instructions for performing the step of analyzing comprise instructions for performing at least one of the following steps:

applying, for each candidate abnormality derived from the standard digital chest image and the first difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network; and

applying, for each candidate abnormality derived from the soft-tissue digital chest image and the second difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network.

15. A system for computerized detection of lung abnormalities, comprising:

a mechanism configured to obtain first and second digital chest images, said first digital chest image comprising a standard digital chest image and said second digital chest image comprising a soft-tissue digital chest image derived from a chest x-ray image in which bony structures are removed by subtraction of a first weighted low energy x-ray exposed image from a second weighted high energy x-ray exposed image;

a mechanism configured to generate a first difference image from the standard digital chest image by a step of signal-to-noise ratio (SNR) suppressing filtering of said standard digital chest image to produce a SNR-suppressed standard image, and a step of SNR enhancing filtering of said standard digital chest image to produce a SNR-enhanced standard image, and a step of producing a difference image between said SNR-suppressed standard image and said SNR-enhanced standard image;

a mechanism configured to generate a second difference image from the soft-tissue digital chest image by a step of signal-to-noise ratio (SNR) suppressing filtering of said soft-tissue digital chest image to produce a SNR-suppressed soft-tissue digital chest image, and a step of SNR enhancing filtering of said soft-tissue digital chest image to produce a SNR-enhanced soft-tissue digital chest image, and a step of producing a difference image between said SNR-suppressed soft-tissue digital chest image and said SNR-enhanced soft-tissue digital chest image;

a mechanism configured to identify candidate abnormalities in the first and second difference images;

a mechanism configured to extract from the standard digital chest image and the first difference image predetermined first features of each of the candidate abnormalities identified in the first difference image;

a mechanism configured to extract from the soft-tissue digital chest image and the second difference images predetermined second features of each of the candidate abnormalities identified in the second difference image;

a mechanism configured to analyze the extracted first features and the extracted second features to identify and eliminate false positive candidate abnormalities respectively corresponding thereto;

a mechanism configured to perform a logical OR operation of the candidate abnormalities derived respectively from the first and second difference images and remaining after the elimination of the false positive candidate abnormalities; and

a mechanism configured to output a signal indicative of a result of performing the logical OR operation.

16. The system of claim 15, wherein the output mechanism is configured to display at least one of a standard image and a soft-tissue image and indicate thereon a location of at least one of:

the candidate abnormalities derived from the first difference image, remaining after the elimination of the false positive candidate abnormalities,

the candidate abnormalities derived from the second difference image, remaining after the elimination of the false positive candidate abnormalities, and

the candidate abnormalities derived from the logical OR operation.

17. The system of claim 15, wherein the analyzing mechanism is configured to analyze the candidate abnormalities using adaptive rule-based analysis on plural of the extracted features.

18. The system of claim 15, wherein the analyzing mechanism comprises:

a mechanism configured to analyze candidate abnormalities identified in said first difference image using adaptive rule-based analysis specific to features extracted from the standard digital chest image and the first difference image to identify and eliminate false positive candidate abnormalities identified in said first difference image; and

a mechanism configured to analyze candidate abnormalities identified in said second difference image using adaptive rule-based analysis specific to features extracted from the soft-tissue digital chest image and the second difference image to identify and eliminate false positive candidate abnormalities identified in said second difference image.

19. The system of claim 15, wherein the analyzing mechanism comprises:

a mechanism configured to apply, for each candidate abnormality derived from the standard digital chest image and the first difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network; and

a mechanism configured to apply, for each candidate abnormality derived from the soft-tissue digital chest image and the second difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network.

20. The system of claim 17, wherein the analyzing mechanism comprises:

a mechanism configured to apply, for each candidate abnormality derived from the standard digital chest image and the first difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network; and

a mechanism configured to apply, for each candidate abnormality derived from the soft-tissue digital chest image and the second difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network.

21. The system of claim 18, wherein the analyzing mechanism comprises:

a mechanism configured to apply, for each candidate abnormality derived from the standard digital chest image and the first difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network; and

a mechanism configured to apply, for each candidate abnormality derived from the soft-tissue digital chest image and the second difference image, plural extracted features extracted from the respective candidate abnormality to a trained artificial neural network and eliminating false positive candidate abnormalities based on an output of the trained artificial neural network.