Automated measurement of anatomical structures in medical imaging
View Patent ↗An automated process classifies MRI or other medical imaging data as belonging to neurological structures. A training part of the process uses a baseline data set with accompanying classification map, as well as an arbitrary number of training data sets, also with accompanying classification maps. Each training data set is registered to the baseline data set using a warp-based automated registration algorithm. The resulting probability map, mean values and covariance matrices are used to classify structures in an image data set using a maximum likelihood criterion.
1. A method for automated measurement of a structure in an image data set, the method comprising:
(a) providing a baseline data set, a plurality of training data sets and classifying maps identifying the structure in the baseline data set and in each of the plurality of training data sets;
(b) registering each of the plurality of training data sets to the baseline data set to produce a first registration result;
(c) from the first registration result, deriving class statistics comprising a mean value, a covariance matrix, and a probability map indicating a probability that the structure is found in each image element in each of the plurality of training data sets;
(d) registering the baseline data set to the image data set to produce a second registration result;
(e) from the second registration result, registering the probability map to the image data set to produce a third registration result; and
(f) from the third registration result and the class statistics derived from the first registration result, locating the structure in the image data set.
2. The method of claim 1 , wherein at least one of the first and second registration results is produced through warp-based automated registration.
3. The method of claim 1 , wherein step (f) is performed using a maximum likelihood criterion.
4. The method of claim 1 , wherein a plurality of said structures are to be located in the image data set and are represented in both the baseline data set and each of the plurality of training data sets.
5. The method of claim 4 , wherein the probability map represents a probability that each of the plurality of structures is found in each image element in each of the plurality of training data sets.
6. The method of claim 5 , wherein step (c) comprises producing, for each of the plurality of structures, a mean value and a covariance matrix.
7. The method of claim 6 , wherein step (f) is performed using a maximum likelihood criterion.
8. The method of claim 7 , wherein step (f) is performed using, as a convergence criterion, a global maximization of a maximum likelihood discriminant function.
9. The method of claim 1 , wherein the structure is a biological structure in a body.
10. The method of claim 9 , wherein the structure is a neurological structure.
11. The method of claim 9 , wherein the image data set is obtained through magnetic resonance imaging.
12. The method of claim 1 , wherein step (f) comprises locating and orienting a shape-based feature extraction algorithm in accordance with the third registration result.
13. A system for automated measurement of a structure in an image data set, the system comprising:
an input for receiving the image data set;
an input for receiving a baseline data set, a plurality of training data sets and classifying maps identifying the structure in the baseline data set and in each of the plurality of training data sets; and
a computing device, in communication with the inputs, for:
(i) registering each of the plurality of training data sets to the baseline data set to produce a first registration result;
(ii) from the first registration result, deriving class statistics comprising mean value, a covariance matrix and a probability map indicating a probability that the structure is found in each image element in each of the plurality of training data sets;
(iii) registering the baseline data set to the image data set to produce a second registration result;
(iv) from the second registration result, registering the probability map to the image data set to produce a third registration result; and
(v) from the third registration result and the class statistics derived from the first registration result, locating the structure in the image data set.
14. The system of claim 13 , wherein the computing device performs at least one of the first and second registration results through warp-based automated registration.
15. The system of claim 13 , wherein the computing device performs step (v) using a maximum likelihood criterion.
16. The system of claim 13 , wherein a plurality of said structures are to be located in the image data set and are represented in both the baseline data set and each of the plurality of training data sets, and wherein the probability map represents a probability that each of the plurality of structures is found in each image element in each of the plurality of training data sets.
17. The system of claim 16 , wherein the computing device produces, for each of the plurality of structures, a mean value and a covariance matrix as part of step (ii).
18. The system of claim 17 , wherein the computing device performs step (v) using a maximum likelihood criterion.
19. The system of claim 18 , wherein the computing device performs step (v) using, as a convergence criterion, a global maximization of a maximum likelihood discriminant function.
20. The system of claim 13 , wherein the computing device performs step (v) by locating and orienting a shape-based feature extraction algorithm in accordance with the third registration result.
21. A method for producing a training and classification data set for automated measurement of a structure in an image data set, the method comprising:
(a) providing a baseline data set, a plurality of training data sets and classifying maps identifying the structure in the baseline data set and in each of the plurality of training data sets;
(b) registering each of the plurality of training data sets to the baseline data set to produce a registration result;
(c) from the registration result, deriving class statistics comprising a mean value, a covariance matrix and a probability map indicating a probability that the structure is found in each image element in each of the plurality of training data sets; and
(d) writing the class statistics to a file.
22. The method of claim 21 , wherein the registration result is produced through warp-based automated registration.
23. The method of claim 21 , wherein a plurality of said structures are to be located in the image data set and are represented in both the baseline data set and each of the plurality of training data sets.
24. The method of claim 23 , wherein the probability map represents a probability that each of the plurality of structures is found in each image element in each of the plurality of training data sets.
25. The method of claim 24 , wherein step (c) comprises producing, for each of the plurality of structures, a mean value and a covariance matrix.
26. The method of claim 21 , wherein the structure is a biological structure in a body.
27. The method of claim 26 , wherein the structure is a neurological structure.
28. The method of claim 21 , wherein step (d) comprises writing the file to a persistent storage medium.
29. A persistent storage medium formed according to the method of claim 28 .
30. A system for producing a training and classification data set for automated measurement of a structure in an image data set, the system comprising:
an input for receiving a baseline data set, a plurality of training data sets and classifying maps identifying the structure in the baseline data set and in each of the plurality of training data sets; and
a computing device, in communication with the input, for:
(i) registering each of the plurality of training data sets to the baseline data set to produce a registration result;
(ii) from the registration result, deriving class statistics comprising a mean value, a covariance matrix and a probability map indicating a probability that the structure is found in each image element in each of the plurality of training data sets; and
(iii) writing the class statistics to a file.
31. The system of claim 30 , wherein the computing device performs step (i) through warp-based automated registration.
32. The system of claim 30 , wherein a plurality of said structures are to be located in the image data set and are represented in both the baseline data set and each of the plurality of training data sets, and wherein the probability map represents a probability that each of the plurality of structures is found in each image element in each of the plurality of training data sets.
33. The system of claim 32 , wherein step (ii) comprises producing, for each of the plurality of structures, a mean value and a covariance matrix.
34. The system of claim 30 , further comprising a drive, in communication with the computing device, for writing the file to persistent storage.
35. A method for automated measurement of a structure in an image data set, the method comprising:
(a) providing a baseline data set and class statistics comprising a mean value, a covariance matrix and a probability map indicating a probability that the structure is found in any given image element;
(b) registering the baseline data set to the image data set to produce a registration result;
(c) from the registration result, registering the probability map to the image data set to produce a further registration result; and
(d) from the further registration result and the class statistics, locating the structure in the image data set.
36. The method of claim 35 , wherein step (d) is performed using a maximum likelihood criterion.
37. The method of claim 35 , wherein a plurality of said structures are to be located in the image data set and are represented in the baseline data set.
38. The method of claim 37 , wherein the probability map represents a probability that each of the plurality of structures is found in any given image element.
39. The method of claim 38 , wherein the probability map comprises, for each of the plurality of structures, a mean value and a covariance matrix.
40. The method of claim 39 , wherein step (d) is performed using a maximum likelihood criterion.
41. The method of claim 40 , wherein step (d) is performed using, as a convergence criterion, a global maximization of a maximum likelihood discriminant function.
42. The method of claim 35 , wherein the structure is a biological structure in a body.
43. The method of claim 42 , wherein the structure is a neurological structure .
44. The method of claim 43 , wherein the image data set is obtained through magnetic resonance imaging.
45. The method of claim 35 , wherein step (d) comprises locating and orienting a shape-based feature extraction algorithm in accordance with the further registration result.
46. A system for automated measurement of a structure in an image data set, the system comprising:
an input for receiving a baseline data set and class statistics comprising a mean value, a covariance matrix and a probability map indicating a probability that the structure is found in any given image element; and
a computing device, in communication with the input, for:
(i) registering the baseline data set to the image data set to produce a registration result;
(ii) from the registration result, registering the probability map to the image data set to produce a further registration result; and
(iii) from the further registration result and the class statistics, locating the structure in the image data set.
47. The system of claim 46 , wherein the computing device performs step (iii) using a maximum likelihood criterion.
48. The system of claim 46 , wherein a plurality of said structures are to be located in the image data set and are represented in the baseline data set, and wherein the probability map represents a probability that each of the plurality of structures is found in any given image element.
49. The system of claim 48 , wherein the probability map comprises, for each of the plurality of structures, a mean value and a covariance matrix.
50. The system of claim 49 , wherein step (iii) is performed using a maximum likelihood criterion.
51. The system of claim 50 , wherein step (iii) is performed using, as a convergence criterion, a global maximization of a maximum likelihood discriminant function.
52. The system of claim 46 , wherein the computing device performs step (iii) by locating and orienting a shape-based feature extraction algorithm in accordance with the further registration result.