METHOD FOR CHECKING THE SEGMENTATION OF A STRUCTURE IN IMAGE DATA

• US 20120134552A1
• Filed: 05/31/2011
• Published: 05/31/2012
• Est. Priority Date: 06/01/2010
• Status: Active Grant

• Current Assignee: Siemens Healthcare GMBH (Siemens AG)
• Original Assignee: Siemens AG
• Inventors: Boettger, Thomas
• Granted Patent: US 8,682,074 B2
• Time in Patent Office: Days
• US Class Current: 382/128
• CPC Class Codes: G06T 2207/30168 Image quality inspection
• CPC Class Codes: G06T 7/12 Edge-based segmentation

Abstract

The present embodiments relate to a method for checking a segmentation of a structure in image data, where the segmentation to be checked provides a segmentation region that approximates the structure imaged in the image data and that is delimited by a segmentation contour. The method includes spatially resolved automatic determination of characteristic values that are suitable for serving as an indicator of the accuracy of the segmentation of the structure on the basis of the image data, the segmentation contour to be checked, or a segmentation method applied during the performance of the segmentation to be checked. The method also includes spatially resolved automatic determination of confidence values for a plurality of positions on the segmentation contour using the characteristic values determined for the corresponding positions. The confidence values may indicate the reliability of the segmentation to be checked at the corresponding positions.

Claims

1. A method for checking a segmentation of a structure in image data, the segmentation to be checked providing a segmentation region that approximates the structure imaged in the image data and is delimited by a segmentation contour, the method comprising:

automatically determining spatially resolved characteristic values serving as an indicator of an accuracy of the segmentation of the structure on the basis of the image data, the segmentation contour to be checked, or a segmentation method applied during the performance of the segmentation to be checked; and

automatically determining spatially resolved confidence values for a plurality of positions on the segmentation contour using the automatically determined characteristic values determined for corresponding positions.

2. The method as claimed in claim 1, wherein the spatially resolved characteristic values are determined on the basis of the image data, and

wherein the spatially resolved characteristic values include gradients of intensity values in the image data, in which the structure is imaged, gradients of probability values of a probability map determined using a random walker segmentation method during a segmentation of the structure in the image data in the course of the method for checking, or curvature values of probability isocontour lines in the probability map.

3. The method as claimed in claim 1, wherein the spatially resolved characteristic values are determined on the basis of the segmentation contour to be checked, and

wherein the spatially resolved characteristic values include curvature values of the segmentation contour to be checked.

4. The method as claimed in claim 1, wherein the spatially resolved characteristic values are determined on the basis of the applied segmentation method, and

wherein the spatially resolved characteristic values include gradients of probability values of a probability map determined using a random walker segmentation method applied during the performance of the segmentation to be checked, curvature values of probability isocontour lines in the probability map, convergence rates for different sections of the segmentation contour to be checked that occur during the performance of the segmentation using the segmentation method, or changes in values of an energy functional determined using the segmentation method with variations of the segmentation contour.

5. The method as claimed in claim 1, wherein the confidence value for a position of the plurality is determined by statistical analysis of the characteristic values that are arranged in a specific region around the position.

6. The method as claimed in claim 5, wherein the statistical analysis includes determining a simple or weighted average value or a median of the characteristic values in the determined region.

7. The method as claimed in claim 1, wherein the confidence values are determined such that higher characteristic values lead to higher confidence values.

8. The method as claimed in claim 1, wherein the segmentation contour in the image data is described by positions of pixels or voxels that lie on the segmentation contour or are intersected by the segmentation contour, a confidence value being determined for each of the pixels or each of the voxels of the segmentation contour.

9. The method as claimed in claim 1, further comprising performing a random walker segmentation method to segment the structure in the image data, probability values for the image data being determined using the random walker segmentation method,

wherein the spatially resolved characteristic values are determined by determining gradients of the probability values.

10. The method as claimed in claim 9, wherein seed points for the random walker segmentation method are determined on the basis of the segmentation to be checked.

11. The method as claimed in claim 9, wherein performing the random walker segmentation method comprises:

automatically determining a signed distance function that indicates distances from the segmentation contour of the segmentation to be checked;

automatically specifying seed points inside and outside the segmentation region for a predetermined distance from the segmentation contour; and

automatically performing the random walker segmentation method using the specified seed points.

12. The method as claimed in claim 1, wherein the segmentation to be checked is produced using a random walker segmentation method, with which probability values for the image data are determined, and

wherein the spatially resolved characteristic values are determined by determining gradients of the probability values.

13. The method as claimed in claim 1, wherein the segmentation to be checked is produced using a random walker segmentation method, and

wherein the spatially resolved characteristic values are determined by determining a curvature of the segmentation contour of the segmentation to be checked.

14. The method as claimed in claim 1, wherein the segmentation of the structure in the image data is continued starting from the segmentation to be checked using the determined confidence values.

15. The method as claimed in claim 1, wherein the spatially resolved characteristic values are determined on the basis of the segmentation method, and

wherein the method further comprises:

automatically modifying a parameter of the segmentation method, using which the segmentation to be checked is produced, on the basis of the determined confidence values; and

re-determining the segmentation using the modified segmentation method.

16. The method as claimed in claim 15, wherein the segmentation to be checked is produced using a random walker segmentation method, and

wherein at least a threshold value for a binary extraction of the segmentation region from a probability map is adjusted locally on the basis of the confidence values.

17. The method as claimed in claim 1, wherein the segmentation to be checked is a manual segmentation performed by user input.

18. The method as claimed in claim 1, further comprising displaying, in a spatially resolved mariner, the confidence values for the segmentation contour.

19. A device for checking a segmentation of a structure in image data, the segmentation to be checked providing a segmentation region that approximates the structure imaged in the image data and is delimited by a segmentation contour, the device comprising:

a computer unit configured to;

automatically determine, in a spatially resolved manner, characteristic values that serve as an indicator of an accuracy of the segmentation of the structure on the basis of the image data, the segmentation contour to be checked, or a segmentation method applied during the performance of the segmentation to be checked; and

automatically determine, in a spatially resolved manner, confidence values for a plurality of positions on the segmentation contour using the characteristic values determined for corresponding positions,

wherein the confidence values indicate a reliability of the segmentation to be checked at the corresponding positions.

20. The device as claimed in claim 19, wherein the spatially resolved characteristic values are determined on the basis of the image data, and

wherein the spatially resolved characteristic values include gradients of intensity values in the image data, in which the structure is imaged, gradients of probability values of a probability map determined using a random walker segmentation method during a segmentation of the structure in the image data in the course of the method for checking, or curvature values of probability isocontour lines in the probability map.

21. In a non-transitory computer readable medium comprising computer readable instructions that, when executed by a computer system, perform a method for checking a segmentation of a structure in image data, the segmentation to be checked providing a segmentation region that approximates the structure imaged in the image data and is delimited by a segmentation contour, the method comprising:

automatically determining characteristic values serving as an indicator of an accuracy of the segmentation of the structure for different positions, the determining being on the basis of the image data, the segmentation contour to be checked, or a segmentation method applied during the performance of the segmentation to be checked; and

automatically determining confidence values for a plurality of the different positions on the segmentation contour using the automatically determined characteristic values.

22. In a non-transitory computer program product comprising a computer program that, when executed by a computer system, performs a method for checking a segmentation of a structure in image data, the segmentation to be checked providing a segmentation region that approximates the structure imaged in the image data and that is delimited by a segmentation contour, the method comprising:

automatically determining, in a spatially resolved manner, characteristic values serving as an indicator of an accuracy of the segmentation of the structure on the basis of the image data, the segmentation contour to be checked, or a segmentation method applied during the performance of the segmentation to be checked; and

automatically determining, in a spatially resolved manner, confidence values for a plurality of positions on the segmentation contour using the automatically determined characteristic values determined for corresponding positions.