Method and system for left ventricle detection in 2D magnetic resonance images
View Patent ↗A method and system for left ventricle (LV) detection in 2D magnetic resonance imaging (MRI) images is disclosed. In order to detect the LV in a 2D MRI image, a plurality of LV candidates are detected, for example using marginal space learning (MSL) based detection. Candidates for distinctive anatomic landmarks associated with the LV are then detected in the 2D MRI image. In particular, apex candidates and base candidates are detected in the 2D MRI image. One of the LV candidates is selected as a final LV detection result using component-based voting based on the detected LV candidates, apex candidates, and base candidates.
1. A method for left ventricle (LV) detection in a 2D magnetic resonance imaging (MRI) image, comprising:
detecting a plurality of LV candidates in the 2D MRI image;
detecting a plurality of apex candidates in the 2D MRI image;
detecting a plurality of base candidates in the 2D MRI image; and
selecting one of the plurality LV candidates using component-based voting based on each of the plurality of LV candidates, the plurality of apex candidates, and the plurality of base candidates, wherein said step of selecting one of the plurality LV candidates using component-based voting based on each of the plurality of LV candidates, the plurality of apex candidates, and the plurality of base candidates comprises:
determining votes for each of the plurality of LV candidates based on other LV candidates of the plurality of LV candidates,
determining votes for each of the plurality of LV candidates based on the apex candidates,
determining votes for each of the plurality of LV candidates based on the base candidates, and
selecting as a detection result the one of the plurality of LV candidates having a largest number of votes.
2. The method of claim 1 , wherein said step of detecting a plurality of LV candidates in the 2D MRI image comprises:
detecting the plurality of LV candidates using marginal space learning (MSL).
3. The method of claim 1 , wherein said step of detecting a plurality of LV candidates in the 2D MRI image comprises:
detecting a plurality of position candidates for the LV in the 2D MRI image using a trained position detector;
generating a plurality of position-orientation hypotheses from each of the plurality of position candidates;
detecting a plurality of position-orientation candidates from the plurality of position-orientation hypotheses using a trained position-orientation detector;
generating a plurality of position-orientation-scale hypotheses from each of the plurality of position-orientation candidates; and
detecting the plurality of LV candidates from the plurality of position-orientation-scale hypotheses using a trained position-orientation-scale detector.
4. The method of claim 3 , wherein each of the position detector, the position-orientation detector, and the position-orientation-scale detector are trained using a probabilistic boosting tree (PBT).
5. The method of claim 2 , wherein the plurality of apex candidates and the plurality of base candidates are detected using MSL.
6. The method of claim 1 , wherein said step of determining votes for each of the plurality of LV candidates based on other LV candidates of the plurality of LV candidates comprises, for each LV candidate:
determining a vote for the LV candidate from each of the other LV candidates located within a certain distance of the LV candidate.
7. The method of claim 1 , wherein said step of determining votes for each of the plurality of LV candidates based on the apex candidates comprises, for each LV candidate:
determining a predicted position of the apex for the LV candidate; and
determining a vote for the LV candidate from each apex candidate having a center point within a certain distance of the predicted position of the apex for the LV candidate.
8. The method of claim 1 , wherein said step of determining votes for each of the plurality of LV candidates based on the base candidates comprises, for each LV candidate:
determining a predicted position of a basal center for the LV candidate; and
determining a vote for the LV candidate from each base candidate having a center point within a certain distance of the predicted position of the basal center for the LV candidate.
9. The method of 1 , further comprising:
determining predicted apex and base positions in the 2D MRI image based on the selected one of the plurality of LV candidates;
detecting the apex of the LV around the predicted apex position in the 2D MRI image; and
detecting the base of the LV around the predicted base position in the 2D MRI image.
10. An apparatus for left ventricle (LV) detection in a 2D magnetic resonance imaging (MRI) image, comprising:
means for detecting a plurality of LV candidates in the 2D MRI image;
means for detecting a plurality of apex candidates in the 2D MRI image;
means for detecting a plurality of base candidates in the 2D MRI image; and
means for selecting one of the plurality LV candidates using component-based voting based on each of the plurality of LV candidates, the plurality of apex candidates, and the plurality of base candidates, wherein said means for selecting one of the plurality LV candidates using component-based voting on based each of the plurality of LV candidates, the plurality of apex candidates, and the plurality of base candidates comprises:
means for determining votes for each of the plurality of LV candidates based on other LV candidates of the plurality of LV candidates,
means for determining votes for each of the plurality of LV candidates based on the apex candidates,
means for determining votes for each of the plurality of LV candidates based on the base candidates, and
means for selecting as a detection result the one of the plurality of LV candidates having a largest number of votes.
11. The apparatus of claim 10 , wherein said means for detecting a plurality of LV candidates in the 2D MRI image comprises:
means for detecting a plurality of position candidates for the LV in the 2D MRI image using a trained position detector;
means for generating a plurality of position-orientation hypotheses from each of the plurality of position candidates;
means for detecting a plurality of position-orientation candidates from the plurality of position-orientation hypotheses using a trained position-orientation detector;
means for generating a plurality of position-orientation-scale hypotheses from each of the plurality of position-orientation candidates; and
means for detecting the plurality of LV candidates from the plurality of position-orientation-scale hypotheses using a trained position-orientation-scale detector.
12. The apparatus of claim 10 , wherein the plurality of LV candidates, the plurality of apex candidates, and the plurality of base candidates are detected using MSL.
13. The apparatus of claim 10 , wherein said means for determining votes for each of the plurality of LV candidates based on other LV candidates of the plurality of LV candidates comprises:
means for determining a vote for a particular LV candidate from each of the other LV candidates located within a certain distance of the particular LV candidate.
14. The apparatus of claim 10 , wherein said means for determining votes for each of the plurality of LV candidates based on the apex candidates comprises:
means for determining a predicted position of the apex for each LV candidate; and
means for determining a vote for a particular LV candidate from each apex candidate having a center point within a certain distance of the predicted position of the apex for the particular LV candidate.
15. The apparatus of claim 10 , wherein said means for determining votes for each of the plurality of LV candidates based on the base candidates comprises:
means for determining a predicted position of a basal center for each LV candidate; and
means for determining a vote for a particular LV candidate from each base candidate having a center point within a certain distance of the predicted position of the basal center for the particular LV candidate.
16. A non-transitory computer readable medium encoded with computer executable instructions for left ventricle (LV) detection in a 2D magnetic resonance imaging (MRI) image, the computer executable instructions defining steps comprising:
detecting a plurality of LV candidates in the 2D MRI image;
detecting a plurality of apex candidates in the 2D MRI image;
detecting a plurality of base candidates in the 2D MRI image; and
selecting one of the plurality LV candidates using component-based voting based on each of the plurality of LV candidates, the plurality of apex candidates, and the plurality of base candidates, wherein the computer executable instructions defining the step of selecting one of the plurality LV candidates using component-based voting based on each of the plurality of LV candidates, the plurality of apex candidates, and the plurality of base candidates comprise computer executable instructions defining the steps of:
determining votes for each of the plurality of LV candidates based on other LV candidates of the plurality of LV candidates,
determining votes for each of the plurality of LV candidates based on the apex candidates,
determining votes for each of the plurality of LV candidates based on the base candidates, and
selecting as a detection result the one of the plurality of LV candidates having a largest number of votes.
17. The non-transitory computer readable medium of claim 16 , wherein the plurality of LV candidates, the plurality of apex candidates, and the plurality of base candidates are detected using MSL.
18. The non-transitory computer readable medium of claim 16 , wherein the computer executable instructions defining the step of detecting a plurality of LV candidates in the 2D MRI image comprise computer executable instructions defining the steps of:
detecting a plurality of position candidates for the LV in the 2D MRI image using a trained position detector;
generating a plurality of position-orientation hypotheses from each of the plurality of position candidates;
detecting a plurality of position-orientation candidates from the plurality of position-orientation hypotheses using a trained position-orientation detector;
generating a plurality of position-orientation-scale hypotheses from each of the plurality of position-orientation candidates; and
detecting the plurality of LV candidates from the plurality of position-orientation-scale hypotheses using a trained position-orientation-scale detector.
19. The non-transitory computer readable medium of claim 16 , wherein the computer executable instructions defining the step of determining votes for each of the plurality of LV candidates based on other LV candidates of the plurality of LV candidates comprise computer executable instructions defining the step of, for each LV candidate:
determining a vote for the LV candidate from each of the other LV candidates located within a certain distance of the LV candidate.
20. The non-transitory computer readable medium of claim 16 , wherein the computer executable instructions defining the step of determining votes for each of the plurality of LV candidates based on the apex candidates comprise computer executable instructions defining the steps of, for each LV candidate:
determining a predicted position of the apex for the LV candidate; and
determining a vote for the LV candidate from each apex candidate having a center point within a certain distance of the predicted position of the apex for the LV candidate.
21. The non-transitory computer readable medium of claim 16 , wherein said the computer executable instructions defining the step of determining votes for each of the plurality of LV candidates based on the base candidates comprise computer executable instructions defining the steps of, for each LV candidate:
determining a predicted position of a basal center for the LV candidate; and
determining a vote for the LV candidate from each base candidate having a center point within a certain distance of the predicted position of the basal center for the LV candidate.