Method and system for nodule feature extraction using background contextual information in chest x-ray images

1. A method for extracting nodule features in a chest x-ray image for detecting false positives in nodule candidates, comprising:

generating a representation of vessel trees in the chest x-ray image, the representation of the vessel trees defining background contextual information in the chest x-ray image;

extracting at least one feature for each of the nodule candidates based on the background contextual information; and

detecting false positives in the nodule candidates based on the at least one feature detected for each of the nodule candidates.

2. The method of claim 1 , further comprising:

detecting the nodule candidates in the chest x-ray image using an automatic nodule detection method.

3. The method of claim 2 , wherein said step of detecting the nodule candidates in the chest x-ray image comprises:

detecting the nodule candidates in the chest x-ray image using a multi-filter based nodule detection method.

4. The method of claim 1 , wherein said step of generating a representation of vessel trees in the chest x-ray imam comprises:

generating initial vessel tree templates in the chest x-ray image;

propagating vessel tree regions from the initial vessel tree templates; and

validating the propagated vessel tree regions using shape analysis.

5. The method of claim 4 , wherein said step of generating initial vessel tree templates comprises:

generating the initial vessel tree templates at low inner boundaries of lung regions in the chest x-ray image.

6. The method of claim 4 , wherein said step of propagating vessel tree regions from the initial vessel tree templates comprises:

propagating vessel tree regions from the initial vessel tree templates at a plurality of different confidence levels.

7. The method of claim 6 , wherein said step of extracting at least one feature for each of the nodule candidates based on the background contextual information comprises, for each nodule candidate:

extracting a first feature for the nodule candidate by calculating a regular shape of an empty region within a region of interest associated with the nodule candidate, said empty region enclosed by vessel trees propagated at a highest confidence level;

extracting a second feature for the nodule candidate by calculating an average propagation distance of vessel tree pixels next to said empty region at the highest confidence level;

extracting a third feature for the nodule candidate by calculating a relative size of said empty region with respect to said region of interest at the highest confidence level; and

extracting a fourth feature for the nodule candidate by calculating a weighted sum of confidence levels of boundary pixels in vessel tree regions that are next to said empty region.

8. The method of claim 1 , wherein said background contextual information is the representation of vessel trees in lung regions of the chest x-ray image, and said step of extracting at least one feature for each of the nodule candidates based on the background contextual information comprises:

extracting at least one feature representing a relationship between each of the nodule candidates and the vessel trees in the chest x-ray image.

9. The method of claim 1 , wherein said step of detecting false positives in the nodule candidates based on the at least one feature detected for each of the nodule candidates comprises:

comparing said at least one feature extracted for each nodule candidate to a threshold to determine whether each nodule candidate is a false positive or a genuine nodule.

10. An apparatus for extracting nodule features in a chest x-ray image for detecting false positives in nodule candidates, comprising:

means for generating a representation of vessel trees in the chest x-ray image, the representation of the vessel trees defining background contextual information in the chest x-ray image;

means for extracting at least one feature for each of the nodule candidates based on the background contextual information; and

means for detecting false positives in the nodule candidates based on the at least one feature detected for each of the nodule candidates.

11. The apparatus of claim 10 , further comprising:

means for detecting the nodule candidates in the chest x-ray image using an automatic nodule detection method.

12. The apparatus of claim 10 , wherein said means for generating are representation of vessel trees in the chest x-ray image comprises:

means for generating initial vessel tree templates in the chest x-ray image;

means for propagating vessel tree regions from the initial vessel tree templates; and

means for validating the propagated vessel tree regions using shape analysis.

13. The apparatus of claim 12 , wherein said means for propagating vessel tree regions from the initial vessel tree templates comprises:

means for propagating vessel tree regions from the initial vessel tree templates at a plurality of different confidence levels.

14. The apparatus of claim 13 , wherein said means for extracting at least one feature for each of the nodule candidates based on the background contextual information comprises:

means for extracting a first feature for a nodule candidate by calculating a regular shape of an empty region within a region of interest associated with the nodule candidate, said empty region enclosed by vessel trees propagated at a highest confidence level;

means for extracting a second feature for the nodule candidate by calculating an average propagation distance of vessel tree pixels next to said empty region at the highest confidence level;

means for extracting a third feature for the nodule candidate by calculating a relative size of said empty region with respect to said region of interest at the highest confidence level; and

means for extracting a fourth feature for the nodule candidate by calculating a weighted sum of confidence levels of boundary pixels in vessel tree regions that are next to said empty region.

15. The apparatus of claim 10 , wherein said background contextual information is the representation of vessel trees in lung regions of the chest x-ray image, and said means for extracting at least one feature for each of the nodule candidates based on the background contextual information comprises:

means for extracting at least one feature representing a relationship between each of the nodule candidates and the vessel trees in the chest x-ray image.

16. The apparatus of claim 10 , wherein said means for detecting false positives in the nodule candidates based on the at least one feature detected for each of the nodule candidates comprises:

means for comparing said at least one feature extracted for each nodule candidate to a threshold to determine whether each nodule candidate is a false positive or a genuine nodule.

17. A computer readable medium encoded with computer executable instructions for extracting nodule features in a chest x-ray image for detecting false positives in nodule candidates, the computer executable instructions defining steps comprising:

generating a representation of vessel trees in the chest x-ray image, the representation of the vessel trees defining background contextual information in the chest x-ray image;

extracting at least one feature for each of the nodule candidates based on the background contextual information; and

detecting false positives in the nodule candidates based on the at least one feature detected for each of the nodule candidates.

18. The computer readable medium of claim 17 , further comprising computer executable instructions defining the step of:

detecting the nodule candidates in the chest x-ray image using an automatic nodule detection method.

19. The computer readable medium of claim 17 , wherein the computer executable instructions defining the step of generating a representation of vessel trees in the chest x-ray image comprise computer executable instructions defining the steps of:

generating initial vessel tree templates in the chest x-ray image;

propagating vessel tree regions from the initial vessel tree templates; and

validating the propagated vessel tree regions using shape analysis.

20. The computer readable medium of claim 19 , wherein the computer executable instructions defining the step of propagating vessel tree regions from the initial vessel tree templates comprise computer executable instructions defining the step of:

propagating vessel tree regions from the initial vessel tree templates at a plurality of different confidence levels.

21. The computer readable medium of claim 20 , wherein the computer executable instructions defining the step of extracting at least one feature for each of the nodule candidates based on the background contextual information comprise computer executable instructions defining the steps of, for each nodule candidate:

extracting a first feature for the nodule candidate by calculating a regular shape of an empty region within a region of interest associated with the nodule candidate, said empty region enclosed by vessel trees propagated at a highest confidence level;

extracting a second feature for the nodule candidate by calculating an average propagation distance of vessel tree pixels next to said empty region at the highest confidence level;

extracting a third feature for the nodule candidate by calculating a relative size of said empty region with respect to said region of interest at the highest confidence level; and

extracting a fourth feature for the nodule candidate by calculating a weighted sum of confidence levels of boundary pixels in vessel tree regions that are next to said empty region.

22. The computer readable medium of claim 17 , wherein said background contextual information is the representation of vessel trees in lung regions of the chest x-ray image, and the computer executable instructions defining the step of extracting at least one feature for each of the nodule candidates based on the background contextual information comprise computer executable instructions defining the step of:

extracting at least one feature representing a relationship between each of the nodule candidates and the vessel trees in the chest x-ray image.

23. The computer readable medium of claim 17 , wherein the computer executable instructions defining the step of detecting false positives in the nodule candidates based on the at least one feature detected for each of the nodule candidates comprise computer executable instructions defining the step of:

comparing said at least one feature extracted for each nodule candidate to a threshold to determine whether each nodule candidate is a false positive or a genuine nodule.