System and method for anatomical structure parsing and detection

1. A method of segmenting a digitized image comprising the steps of:

providing a digitized image comprising a plurality of intensities defined on a lattice of pixels;

initializing a Bayesian prior probability and likelihood function for segmenting said image;

decomposing said image into a union of a plurality of subspaces wherein each subspace is a product of k-partitions, and distributing a plurality Bayesian posterior probability functions over said subspaces;

updating said Bayesian posterior probability functions via a data-driven Markov-chain; and

selecting a subset of said Bayesian posterior probability functions that approximate a true Bayesian posterior probability function.

2. The method of claim 1 , wherein updating said Bayesian posterior probability functions further comprises sampling moves from a proposal probability and accepting moves based on a proposal likelihood probability function.

3. The method of claim 1 , wherein said Markov-chain comprises jump moves and diffusion processes.

4. The method of claim 1 , wherein said data-driven Markov-chain combines bottom-up proposals and top-down generative models to guide the updating of the posterior probability function.

5. The method of claim 4 , wherein the bottom proposals are derived from an AdaBoost classifier algorithm.

6. The method of claim 5 , wherein the AdaBoost algorithm outputs a conditional probability.

7. The method of claim 5 , wherein the generative models are based on models using parameters estimated from training said AdaBoost classifiers.

8. The method of claim 1 , further comprising computing an integral volume from said image to compute Haar type features.

9. The method of claim 5 , wherein said AdaBoost classifier uses DOG and DOOG filters to define features.

10. The method of claim 1 , further comprising using an occlusion mask to correct a feature value for an occluded feature, wherein said occlusion mask excludes invalid pixels from a feature value calculation.

11. A program storage device readable by a computer, tangibly embodying a program of instructions executable by the computer to perform the method steps for segmenting a digitized image, said method comprising the steps of:

providing a digitized image comprising a plurality of intensities defined on a lattice of pixels;

initializing a Bayesian prior probability and likelihood function for segmenting said image;

decomposing said image into a union of a plurality of subspaces wherein each subspace is a product of k-partitions, and distributing a plurality Bayesian posterior probability functions over said subspaces;

updating said Bayesian posterior probability functions via a data-driven Markov-chain; and

selecting a subset of said Bayesian posterior probability functions that approximate a true Bayesian posterior probability function.

12. The computer readable program storage device of claim 11 , wherein updating said Bayesian posterior probability functions further comprises sampling moves from a proposal probability and accepting moves based on a proposal likelihood probability function.

13. The computer readable program storage device of claim 11 , wherein said Markov-chain comprises jump moves and diffusion processes.

14. The computer readable program storage device of claim 11 , wherein said data-driven Markov-chain combines bottom-up proposals and top-down generative models to guide the updating of the posterior probability function.

15. The computer readable program storage device of claim 14 , wherein the bottom proposals are derived from an AdaBoost classifier algorithm.

16. The computer readable program storage device of claim 15 , wherein the AdaBoost algorithm outputs a conditional probability.

17. The computer readable program storage device of claim 15 , wherein the generative models are based on models using parameters estimated from training said AdaBoost classifiers.

18. The computer readable program storage device of claim 11 , further comprising computing an integral volume from said image to compute Haar type features.

19. The computer readable program storage device of claim 15 , wherein said AdaBoost classifier uses DOG and DOOG filters to define features.

20. The computer readable program storage device of claim 11 , further comprising using an occlusion mask to correct a feature value for an occluded feature, wherein said occlusion mask excludes invalid pixels from a feature value calculation.

21. A method of segmenting a digitized image comprising the steps of:

providing a digitized image comprising a plurality of intensities defined on a lattice of pixels;

training an AdaBoost classifier that outputs conditional probabilities, and using parameters estimated from said training to define generative models for structures sought to be segmented;

using a data-driven Markov-chain to estimate a plurality of Bayesian posterior probability functions that segment said image, wherein said Markov-chain combines proposals derived from said AdaBoost classifier and said generative models to guide the updating of the posterior probability functions; and

selecting a subset of said Bayesian posterior probability functions that approximate a true Bayesian posterior probability function for the image segmentation.

22. The method of claim 21 , wherein said Markov-chain is ergodic.

23. The method of claim 21 , wherein selecting a subset of said Bayesian posterior probability functions further comprises seeking those functions that minimize a Kullback-Leibler divergence from said approximate posterior probability to a true posterior probability function.