METHOD AND APPARATUS FOR CLASSIFICATION OF CORONARY ARTERY IMAGE DATA

• US 20100082692A1
• Filed: 09/24/2008
• Published: 04/01/2010
• Est. Priority Date: 09/24/2008
• Status: Active Grant

• Current Assignee: Toshiba Medical Systems Europe BV (Toshiba Corporation)
• Original Assignee: Toshiba Medical Systems Europe BV (Toshiba Corporation)
• Inventors: Murphy, Sean, Poole, Ian, Akinyemi, Akinola
• Granted Patent: US 7,941,462 B2
• Time in Patent Office: Days
• US Class Current: 707/797
• CPC Class Codes: G06F 18/24155 Bayesian classification
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers
• CPC Class Codes: G06T 2207/30101 Blood vessel; Artery; Vein; Vascular
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06T 7/162 involving graph-based methods
• CPC Class Codes: G06V 10/426 Graphical representations
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

A polyline tree representation of a coronary artery tree imaged in a volume data set is obtained, and its topology is extracted to give a topological representation indicating the relative positions of vessels in the tree. The topological representation is compared with a set of topological rules to find possible anatomical classifications for each vessel, and a set of candidate labeled polyline trees is generated by labeling the polyline tree with labels showing each combination of possible anatomical classifications. Each candidate labeled tree is filtered according to a set of geometric rules pertaining to spatial characteristics of vessels in arterial trees, and any labeled tree not satisfying the geometric rules is rejected A figure of merit is calculated for each remaining candidate by comparing features of the vessels measured from the polyline tree and from the volume data set with features of correctly classified vessels in other data sets to determine the probable correctness of the labeling of each candidate, and the candidate with the best figure of merit is selected as showing the proper classification of the vessels.

Claims

1. A method of automatically classifying anatomical features shown in a medical image volume data set, the method comprising:

obtaining a polyline tree comprising a plurality of connected points in the data set corresponding to the centrelines of vessels in an arterial tree imaged in the data set, each vessel in the arterial tree being represented by a segment in the polyline tree;

forming a topological representation of the polyline tree which indicates the relative generational positions of the segments within the polyline tree;

comparing the topological representation with a set of topological rules specifying anatomically permissible relative generational positions of vessels in an arterial tree to identify feasible anatomical classifications for the vessels represented by each segment in the polyline tree;

generating a set of candidate labeled polyline trees by associating labels representing the identified anatomical classifications with the corresponding segments in the polyline tree, each candidate labeled polyline tree being one combination of the identified feasible anatomical classifications;

comparing each candidate labeled polyline tree with a set of geometric rules specifying anatomically permissible spatial attributes of vessels in an arterial tree which can be determined from a polyline representation and rejecting any candidate having one or more labels representing vessels which do not comply with the geometric rules;

calculating a figure of merit for each remaining candidate labeled polyline tree by comparing features of the vessels represented in the polyline tree with known features of vessels in the anatomical classes indicated by the labels associated with the segments representing the vessels to determine a probability of the correctness of the labels in each candidate, the figure of merit reflecting the probability; and

identifying the candidate labeled polyline tree having the best figure of merit.

2. A method according to claim 1, in which the topological representation comprises a predecessor array.

3. A method according to claim 1, further comprising associating the labels from the identified candidate with the corresponding vessels in the imaged arterial tree to indicate the classification of the vessels.

4. A method according to claim 1, in which calculating a figure of merit for each remaining candidate labeled polyline tree comprises:

measuring a set of preselected features of each vessel represented in the polyline tree;

combining the measurements for each vessel to calculate a feature vector for each vessel; and

for each candidate labeled polyline tree, comparing the calculated feature vector for each vessel with one or more predetermined parameters derived from features of the anatomical class of vessel indicated by the label associated with the segment representing that vessel to determine a probability that each label is correct, and combining the probabilities within each candidate labeled polyline tree to produce a figure of merit for each candidate.

5. A method according to claim 4, in which the predetermined parameters comprise a mean feature vector and a variance-covariance matrix for the anatomical class of vessel.

6. A method according to claim 5, in which comparing the calculated feature vector for each vessel with a predetermined mean feature vector and a variance-covariance matrix to determine a probability comprises calculating a log-likelihood.

7. A method according to claim 6, in which combining the probabilities to produce a figure of merit comprises summing the log-likelihoods for every labeled segment in the candidate labeled polyline tree.

8. A method according to claim 5, in which the predetermined mean feature vectors are the mean values of a plurality of feature vectors calculated for correctly classified vessels in arterial trees imaged in other data sets and the variance-covariance matrices are calculated for correctly classified vessels in arterial trees imaged in the other data sets.

9. A method according to claim 4, in which comparing the calculated feature vector for each vessel with one or more parameters derived from features of the anatomical class of vessel indicated by the label associated with the segment representing the vessel comprises use of a K-nearest neighbour technique.

10. A method according to claim 4, in which comparing the calculated feature vector for each vessel with one or more parameters derived from features of the anatomical class of vessel indicated by the label associated with the segment representing the vessel comprises use of Baysian histograms.

11. A method according to claim 4, in which measuring the preselected features for each vessel comprises measuring features from the polyline tree and from the data set.

24. A medium storing a computer program product operable to cause a computer system to implement the method according to claim 1.

12. A system for automatically classifying anatomical features shown in a medical image volume data set, the system comprising:

a topology extractor module operable to receive a polyline tree comprising a plurality of connected points in the data set corresponding to the centrelines of vessels in an arterial tree imaged in the data set, each vessel in the arterial tree being represented by a segment in the polyline tree, and to form a topological representation of the polyline tree which indicates the relative generational positions of the segments within the polyline tree;

a set of topological rules specifying anatomically permissible relative generational positions of vessels in an arterial tree;

a set of labels representing anatomical classifications of arterial vessels;

a topology checker module operable to compare the topological representation with the set of topological rules to identify feasible anatomical classifications for the vessels represented by each segment in the polyline tree and to generate a set of candidate labeled polyline trees by associating labels from the label set which represent the identified anatomical classifications with the corresponding segments in the polyline tree, each candidate labeled polyline tree being one combination of the identified feasible anatomical classifications;

a set of geometric rules specifying anatomically permissible spatial attributes of vessels in an arterial tree which can be determined from a polyline representation;

a geometry checker module operable to compare each candidate labeled polyline tree with the set of geometric rules and reject any candidate having one or more labels representing vessels which do not comply with the geometric rules; and

a merit figure module operable to calculate a figure of merit for each remaining candidate labeled polyline tree by comparing features of the vessels represented in the polyline tree with known features of vessels in the anatomical classes indicated by the labels associated with the segments representing the vessels to determine a probability of the correctness of the labels in each candidate, the figure of merit reflecting the probability, and to identify the candidate labeled polyline tree having the best figure of merit.

13. A system according to claim 12, in which the topological representation comprises a predecessor array.

14. A system according to claim 12 in which the merit figure module is further operable to associate the labels from the identified candidate with the corresponding vessels in the imaged arterial tree to indicate the classification of the vessels.

15. A system according to claim 12, further comprising a set of predetermined parameters derived from features of anatomical classes of vessel, calculated from measurements of features of correctly classified vessels in arterial trees in image volume data sets;

and in which;

the merit figure module is operable to calculate a figure of merit for each remaining candidate labeled polyline tree by;

measuring a set of preselected features of each vessel represented in the polyline tree;

combining the measurements for each vessel to calculate a feature vector for each vessel; and

for each candidate labeled polyline tree, comparing the calculated feature vector for each vessel with one or more of the parameters for the anatomical class of vessel indicated by the label associated with the segment representing that vessel to determine a probability that each label is correct, and combining the probabilities within each candidate labeled polyline tree to produce a figure of merit for each candidate.

16. A system according to claim 15, in which the predetermined parameters comprise a mean feature vector and a variance-covariance matrix for each anatomical class of vessel.

17. A system according to claim 16, in which the merit figure module is operable to compare the calculated feature vector for each vessel with a mean feature vector and variance-covariance matrix to determine a probability by calculating a log-likelihood.

18. A system according to claim 17, in which the merit figure module is operable to combine the probabilities to produce a figure of merit by summing the log-likelihoods for every segment in the candidate labeled polyline tree.

19. A system according to claim 16, in which the mean feature vectors are the mean values of a plurality of feature vectors calculated for correctly classified vessels in arterial trees imaged in a plurality of volume data sets and the variance-covariance matrices are calculated for correctly classified vessels in arterial trees imaged in the plurality of volume data sets.

20. A system according to claim 15, in which the merit figure module compares the calculated feature vector for each vessel with one or more of the parameters for the indicated anatomical class of vessel by use of a K-nearest neighbour technique.

21. A system according to claim 15, in which the merit figure module compares the calculated feature vector for each vessel with one or more of the parameters for the indicated anatomical class of vessel by use of Baysian histograms.

22. A system according to claim 15, in which the preselected features for each vessel are measured from the polyline tree and/or from the data set.

23. A system according to claim 16, further comprising:

a feature extractor and classifier module operable to;

measure sets of preselected features of correctly classified vessels in a plurality of polyline trees representing a plurality of imaged arterial trees;

combine the measurements for each vessel to calculate a feature vector for each vessel; and

calculate a mean feature vector and a variance-covariance matrix for each anatomical class of vessel from the calculated feature vectors; and

memory operable to store the mean feature vectors and variance-covariance matrices for retrieval by the merit figure module when calculating figures of merit.