Medical image processing

1. A method for detecting an object in a three-dimensional medical image, comprising:

determining, by a processing device, from data representing the three-dimensional medical image, for each voxel in a portion of the three-dimensional medical image, values of a plurality of features at the voxel, wherein each feature at the voxel characterizes a respective property of the three-dimensional medical image at the voxel;

calculating, by the processing device, a likelihood probability distribution of each feature based on the determined values of the feature and a parameter value derived from training data;

generating, by the processing device, a probability map that indicates a probability of each voxel in the portion of the three-dimensional medical image containing an object to be detected, wherein said generating a probability map includes combining, using Bayes' law, the likelihood probability distributions of the plurality of features at each respective voxel of the portion of the three-dimensional medical image;

detecting, by the processing device, the object in the portion of the three-dimensional medical image based on the probability map; and

outputting, by the processing device, an indication of the detected object.

2. The method of claim 1 , wherein said calculating a likelihood probability distribution of a feature includes selecting a parameter value that generates a predetermined number of false positives in response to performing the method on the training data.

3. The method of claim 1 , wherein the plurality of features includes an appearance feature.

4. The method of claim 3 , wherein the appearance feature comprises image intensity information derived from the three-dimensional medical image.

5. The method of claim 3 , wherein the appearance feature comprises a wavelet feature derived from the three-dimensional medical image.

6. The method of claim 3 , wherein the appearance feature comprises a texture feature derived from the three-dimensional medical image.

7. The method of claim 1 , wherein the plurality of features includes a shape feature.

8. The method of claim 7 , wherein the shape feature includes a first-order shape feature derived from the three-dimensional medical image.

9. The method of claim 7 , wherein the shape feature includes a second-order shape feature.

10. The method of claim 9 , wherein the second-order shape feature is determined from a Hessian matrix.

11. The method of claim 10 , wherein the second-order shape feature is calculated from eigenvalues of the Hessian matrix.

12. The method of claim 9 , wherein the second-order shape feature represents a volumetric shape index of the three-dimensional medical image at a respective voxel.

13. The method of claim 9 , wherein the second-order shape feature represents a second principal curvature flow of the three-dimensional medical image at a respective voxel.

14. The method of claim 1 , further comprising calculating a spatial prior probability.

15. The method of claim 14 , wherein said generating a probability map further comprises combining, using Bayes' law, the likelihood probability distributions and the spatial prior probability.

16. The method of claim 15 , wherein the spatial prior probability comprises a spatial constraint.

17. The method of claim 16 , wherein the spatial constraint is imposed by a Markov Random Field and a Gibbs Random Field.

18. The method of claim 1 , wherein calculating a likelihood probability distribution of a feature at a voxel comprises calculating a value of a Gaussian function at the voxel, and wherein the Gaussian function models the probability distribution of the feature.

19. The method of claim 18 , wherein the Gaussian function models the probability distribution of a feature that characterises the intensity of the three-dimensional medical image, and wherein said calculating a value of a Gaussian function includes treating a mean intensity in the Gaussian function as a fixed value.

20. The method of claim 18 , wherein the Gaussian function models the probability distribution of a feature that characterises the intensity of the three-dimensional medical image, and wherein said calculating a value of a Gaussian function includes calculating a mean intensity in the Gaussian function as a function of a size of a previously-detected object.

21. The method of claim 20 , wherein said calculating a mean intensity in the Gaussian function as a function of a size of a previously-detected object includes calculating an expression that is based upon the presence of a partial volume effect in the three-dimensional medical image.

22. The method of claim 1 , wherein the plurality of features includes an anatomical feature.

23. The method of claim 22 , further comprising determining a value of an anatomical feature, wherein said determining a value of an anatomical feature includes calculating a colon centreline.

24. The method of claim 22 , wherein said determining a value of an anatomical feature includes calculating a colon boundary distance transform.

25. The method of claim 1 , wherein said detecting the object in the three-dimensional medical image comprises performing thresholding on the probability map and labelling regions in the thresholded probability map.

26. The method of claim 1 , further comprising defining the portion of the three-dimensional medical image using a mask.

27. The method of claim 1 , wherein said detecting the object comprises detecting a region within the portion of the three-dimensional medical image having an intensity and a size that are within a predetermined intensity range and size range of a type of target object.

28. The method of claim 1 , wherein the method is used in a candidate generation step of a computer aided detection algorithm.

29. The method of claim 1 , wherein said detecting the object in the three-dimensional medical image includes determining from the probability map whether a previously-detected object is a false positive.

30. The method of claim 1 , further comprising segmenting the three-dimensional medical image based upon a result of detecting the object in the portion of the three-dimensional medical image.

31. A computer-readable storage medium having stored thereon computer-executable instructions, execution of which by a processing device causes the processing device to perform operations for detecting an object in a three-dimensional medical image, the operations comprising:

determining, from data representing the three-dimensional medical image, for each voxel in a portion of the three-dimensional medical image, values of a plurality of features at the voxel, wherein each feature at the voxel characterizes a respective property of the three-dimensional medical image at the voxel;

calculating a likelihood probability distribution of each feature based on the determined value of the feature and a parameter value derived from training data;

generating a probability map that indicates a probability of each voxel in the portion of the three-dimensional medical image containing an object to be detected, wherein said generating a probability map includes combining, using Bayes' law, the likelihood probability distributions of the plurality of features at each respective voxel of the portion of the three-dimensional medical image;

detecting the object in the portion of the three-dimensional medical image based on the probability map; and

outputting an indication of the detected object.

32. An apparatus for detecting an object in a three-dimensional medical image, the apparatus comprising:

a processing device configured to:

determine, from data representing the three-dimensional medical image, for each voxel in a portion of the three-dimensional medical image, values of a plurality of features at the voxel, wherein each feature at the voxel characterizes a respective property of the three-dimensional medical image at the voxel;

calculate a likelihood probability distribution of each feature based on the determined values of the feature and a parameter value derived from training data;

generate a probability map that indicates a probability of each voxel in the portion of the three-dimensional medical image containing an object to be detected, wherein the generating a probability map includes combining, using Bayes' law, the likelihood probability distributions of the plurality of features at each respective voxel of the portion of the three-dimensional medical image;

detect an object in the portion of the three-dimensional medical image based on the probability map; and

output an indication of the detected object.