Automatic cardiac view classification of echocardiography

1. A non-transitory computer readable storage medium embodying instructions executed by a processor to perform a method for view classification, the method comprising:

deriving a region of interest localizing an object of interest within an image for each of a plurality of detectors, wherein each detector is configured to detect a portion of the object of interest from a certain view, and each detector is configured for a different view of the object of interest;

determining a global template for each of the regions of interest, wherein each global template is a bounding box larger than and including the region of interest, wherein the global template corresponds to a cropped portion of the image;

performing a global view classification of the cropped portion of the image using a multiview classifier, wherein the multiview classifier outputs a classification for each of the global templates; and

fusing the classifications of the global templates to determine a bounding box for the object of interest in the image.

2. The computer readable storage medium of claim 1 , wherein the object of interest is a heart.

3. The computer readable storage medium of claim 1 , wherein the region of interest includes a left ventricle of a heart.

4. The computer readable storage medium of claim 3 , wherein the image is an end-diastolic frame of a heart.

5. The computer readable storage medium of claim 1 , wherein the detectors are binary classifiers.

6. The computer readable storage medium of claim 1 , wherein detecting the region of interest comprises:

providing training images for each view; and

training a binary classifier for each view.

7. The computer readable storage medium of claim 1 , wherein the multiview classifier includes response functions, the response functions sharing weak classifiers that are weighted differently for each of the views, wherein each weak classifier is associated with a local image filter.

8. A system for view classification comprising:

a memory device storing a plurality of instructions embodying a multiview classifier;

a processor for receiving an image frame including an object of interest to be classified and executing the multiview classifier to perform a method comprising:

deriving a region of interest localizing the object of interest within an image for each of a plurality of binary classifiers, wherein each binary classifier is configured to detect a portion of the object of interest from a certain view, and each binary classifier is configured for a different view of the object of interest;

determining a global template for each of the regions of interest, wherein each global template is a bounding box larger than and including a respective region of interest, wherein the global template corresponds to a cropped portion of the image;

performing a global view classification of the cropped portion of the image using a multiview classifier, wherein the multiview classifier outputs a classification for each of the global templates; and

fusing the classifications of the global templates to determine a bounding box for the object of interest in the image.

9. The system of claim 8 , wherein the object of interest is a heart.

10. The system of claim 8 , wherein the region of interest includes a left ventricle of a heart.

11. The system of claim 8 , wherein the image is an end-diastolic frame of a heart.

12. The system of claim 8 , wherein detecting the region of interest comprises:

providing training images for each view; and

training a binary classifier for each view.

13. The system of claim 8 , wherein the multiview classifier includes response functions, the response functions sharing weak classifiers that are weighted differently for each of the views, wherein each weak classifier is associated with a local image filter.

14. A method for view classification of a heart, the method comprising:

deriving a region of interest localizing a heart within an end-diastolic image for each of a plurality of detectors, wherein each detector is configured to detect a portion of the heart from a certain view, and each detector is configured for a different view of the heart;

determining a global template for each of the regions of interest, wherein each global template is a bounding box larger than and including a respective region of interest, wherein the global template corresponds to a cropped portion of the end-diastolic image;

performing a global view classification of the cropped portion of the end-diastolic image using a multiview classifier, wherein the multiview classifier outputs a classification for each of the global templates;

fusing the classifications of the global templates to determine a bounding box for the heart in the end-diastolic image; and

determining measurements of the heart based on the bounding box for the heart.

15. The method of claim 1 , wherein the multiview classifier includes response functions, the response functions sharing weak classifiers that are weighted differently for each of the views, wherein each weak classifier is associated with a local image filter.

16. The computer readable storage medium of claim 1 , wherein the global template is determined as having a maximum detector score given the region of interest for a respective detector.

17. The system of claim 8 , wherein the global template is determined as having a maximum detector score given the region of interest for a respective detector.

18. The method of claim 14 , wherein the global template is determined as having a maximum detector score given the region of interest for a respective detector.