Generating an anatomical model using a rule-based segmentation and classification process

• US 20080292153A1
• Filed: 05/25/2007
• Published: 11/27/2008
• Est. Priority Date: 05/25/2007
• Status: Active Grant

• Current Assignee: Definiens AG (Astrazeneca PLC)
• Original Assignee: Definiens AG (Astrazeneca PLC)
• Inventors: Schmidt, Guenter, Binnig, Gerd
• Granted Patent: US 8,989,468 B2
• Time in Patent Office: Days
• US Class Current: 382/128
• CPC Class Codes: G06F 18/241 relating to the classification model, e.g. parametric or non-parametric approaches
• CPC Class Codes: G06T 2200/24 involving graphical user interfaces [GUIs]
• CPC Class Codes: G06T 2207/20108 Interactive selection of 2D slice in a 3D data set
• CPC Class Codes: G06T 2207/30004 Biomedical image processing
• CPC Class Codes: G06T 7/0012 Biomedical image inspection

Abstract

A system for computer-aided detection uses a computer-implemented network structure to analyze patterns present in digital image slices of a human body and to generate a three-dimensional anatomical model of a patient. The anatomical model is generated by detecting easily identifiable organs first and then using those organs as context objects to detect other organs. A user specifies membership functions that define which objects of the network structure belong to the various classes of human organs specified in a class hierarchy. A membership function of a potentially matching class determines whether a candidate object of the network structure belongs to the potential class based on the relation between a property of the voxels linked to the candidate object and a property of the context object. Some voxel properties used to classify an object are location, brightness and volume. The human organs are then measured to assist in the patient'"'"'s diagnosis.

Claims

1. A method comprising:

specifying a first class of a class network, wherein the first class is associated with a first anatomical object;

specifying a second class of the class network, wherein the second class is associated with a second anatomical object, and wherein the second class is specified according to a relation between objects in the second class and objects in the first class;

specifying a process hierarchy, wherein the process hierarchy includes a first process step and a second process step;

performing the first process step to detect the first anatomical object using the first class;

performing the second process step to detect the second anatomical object using the second class, wherein the second process step detects the second anatomical object using the detection of the first anatomical object; and

measuring the second anatomical object.

2. The method of claim 1, wherein the second process step detects the second anatomical object in a three-dimensional dataset of voxels.

3. The method of claim 2, wherein the measuring the second anatomical object measures a mean brightness of voxels that depict the second anatomical object.

4. The method of claim 1, wherein the first anatomical object is an aorta and the second anatomical object is a spine.

5. The method of claim 1, wherein the first anatomical object is a human kidney and the second anatomical object is a human pelvis.

6. The method of claim 1, wherein the second process step dynamically selects the first anatomical object such that the first anatomical object has been detected before the second process step is performed.

7. The method of claim 1, wherein the measuring the second anatomical object measures a volume of the second anatomical object.

8. The method of claim 1, wherein the measuring the second anatomical object measures a change in shape over time of the second anatomical object.

9. The method of claim 8, wherein the second anatomical object is a human liver.

10. The method of claim 1, further comprising:

performing the third process step to detect a sub-object within the second anatomical object; and

measuring the sub-object.

11. The method of claim 10, wherein the measuring the second anatomical object measures a shape of the sub-object, and wherein the shape is measured using a bounding box having x, y and z dimensions.

12. The method of claim 1, wherein the measuring the second anatomical object measures a surface area of the second anatomical object.

13. A computer-implemented network structure, comprising:

a data network including a first set of voxel values, a second set of voxel values, a first object and a second object, wherein the first set of voxel values and the second set of voxel values are part of a digital image slice of a human body, wherein each of the first set of voxel values is linked to the first object, wherein each of the second set of voxel values is linked to the second object, wherein the first object exhibits a first property that depends on the first set of voxel values, and wherein the second object exhibits a second property that depends on the second set of voxel values;

a class network including a first class and a second class, wherein a first membership function associated with the first class determines that the first object belongs to the first class, wherein a second membership function associated with the second class determines that the second object belongs to the second class, wherein the second membership function determines that the second object belongs to the second class based on a relation between the first property and the second property; and

a process hierarchy including a first process step and a second process step, wherein the first process step designates the first class, wherein the second process step designates the second class and has an algorithm, wherein the first process step is performed before the second process step is performed, and wherein the algorithm measures the second property.

14. The network structure of claim 13, wherein the process hierarchy comprises a plurality of process steps including the first process step and the second process step, and wherein an anatomical model is generated when the plurality of process steps are performed.

15. The network structure of claim 13, wherein the first class corresponds to voxels depicting a kidney, and wherein the second class corresponds to voxels depicting a pelvis.

16. The network structure of claim 13, wherein the first membership function is evaluated when the first step is performed.

17. The network structure of claim 13, wherein the second property is a two-dimensional area of voxels that are linked to the second object.

18. The network structure of claim 13, further comprising:

a file system, wherein the second property is stored in the file system.

19. A computer-implemented network structure, comprising:

a data network including a first voxel value, a second voxel value, a first link, a second link, a first object and a second object, wherein the first voxel value and the second voxel value are part of a digital image slice of a human body, wherein the first link links the first voxel value to the first object, and wherein the second link links the second voxel value to the second object;

a class network including a first class and a second class; and

a process hierarchy including a first process step and a second process step, wherein the first process step determines that the first object belongs to the first class, wherein the second process step determines that the second object belongs to the second class based on a relation between the second object and the first object, wherein the first process step is performed before the second process step is performed, wherein the second process step has an algorithm, and wherein the algorithm measures a property of voxels that belong to the second class.

20. The network structure of claim 19, wherein the relation between the second object and the first object is a spatial relation.

21. The network structure of claim 20, wherein the first class corresponds to voxels depicting lungs, wherein the second class corresponds to voxels depicting a liver, and wherein the spatial relation is a relation between a first center voxel of the voxels depicting lungs and a second center voxel of the voxels depicting the liver.

22. The network structure of claim 19, wherein the second process step defines a region of interest based on the first object, and wherein the second process step determines that the second object belongs to the second class based on whether the voxels that belong to the second object are within the region of interest.

23. A system comprising:

a display on which a digital image of a cross section of a human body is displayed; and

means for identifying a first anatomical object and a second anatomical object in the digital image based on a process hierarchy of process steps, wherein the process hierarchy specifies that a first process step identifies the first anatomical object before a second process step identifies the second anatomical object, and wherein the second anatomical object is identified based on the identification of the first anatomical object.

24. The system of claim 23, wherein the means displays an indication of the second anatomical object on the display.

25. The system of claim 24, wherein the indication of the second anatomical object is an outline around the second anatomical object.

26. The system of claim 23, wherein the means identifies a third anatomical object after identifying the second anatomical object, and wherein the first anatomical object is a lung, the second anatomical object is subcutaneous fat, and the third anatomical object is a lymph node.

27. A system comprising:

a display on which a digital image of a cross section of a human body is displayed; and

means for generating an anatomical model of the human body, wherein the anatomical model includes a first organ and a second organ, wherein the means identifies the first organ in the digital image, and wherein the means identifies the second organ based on the identification of the first organ.

28. The system of claim 27, wherein the means displays an indication of the second organ on the display.

29. The system of claim 27, wherein the first organ is a liver, and wherein the second organ is a lymph node.

30. A method comprising:

specifying a first class of a class network;

specifying a second class of the class network;

specifying a third class of the class network, wherein objects of an object network are classified by a first membership function as belonging to the first class, wherein objects of the object network are classified by a second membership function as belonging to the second class, and wherein the first class is associated with a first human organ, the second class is associated with a second human organ, and the third class is associated with a third human organ,

specifying a process hierarchy, wherein the process hierarchy includes a first process step, a second process step and a third process step;

performing the first process step to detect the first human organ using the first membership function;

performing the second process step to detect the second human organ using the second membership function; and

performing the third process step to detect the third human organ using the third membership function, wherein the third membership function classifies objects of the object network as belonging to the third class based on either a first relation between objects in the third class and objects in the first class or based on a second relation between objects in the third class and objects in the second class, and wherein the third membership function classifies objects based on the second relation when the second membership function determines a better fit for the objects classified as belonging to the second class than a fit determined by the first membership function for the objects classified as belonging to the first class.

31. The method of claim 30, further comprising:

acquiring a digital image slice;

superimposing an indication of the third human organ on the digital image slice; and

displaying the indication of the third human organ on a display.

32. The method of claim 30, wherein the digital image slice is one of a plurality of digital image slices that make up a three-dimensional dataset of voxels.

33. The method of claim 30, wherein the first human organ is an aorta and the second human organ is a spine.