Method of segmenting anatomic entities in 3D digital medical images

• US 9,129,390 B2
• Filed: 06/10/2008
• Issued: 09/08/2015
• Est. Priority Date: 06/19/2007
• Status: Active Grant

• Current Assignee: Agfa Nv (Agfa-Gevaert N.V.)
• Original Assignee: AGFA Healthcare NV (Agfa-Gevaert N.V.)
• Inventors: Dewaele, Piet
• Primary Examiner(s): Bella, Matthew
• Assistant Examiner(s): Rosario, Dennis
• Application Number:
  US12/665,658
• Publication Number:
  US 20100189320A1
• Time in Patent Office: 2,646 Days
• Field of Search: 382/173
• US Class Current: 1/1
• CPC Class Codes: G06T 2207/10072 Tomographic images
• CPC Class Codes: G06T 2207/30004 Biomedical image processing
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/162 involving graph-based methods
• CPC Class Codes: G06T 7/168 involving transform domain methods

Abstract

For each of a number of landmarks in an image an initial position of the landmark is defined. Next a neighborhood around the initial position comprising a number of candidate locations of the landmark, is sampled and a cost is associated with each of the candidate locations. A cost function expressing a weighted sum of overall gray level cost and overall shape cost for all candidate locations is optimized. A segmented anatomic entity is defined as a path through a selected combination of candidate locations for which combination the cost function is optimized. During optimization towards the optimal segmented surface/volume graph traversal methods are exploited.

Claims

1. A method of segmenting an anatomic entity in a digital medical image, the method comprising the steps of:

defining, for each of a number of landmark points in the digital medical image, an initial position of each of the landmark points;

connecting a plurality of pairs of the landmark points to each other to form polygons defining a mesh;

constructing a rectangular grid around each of the landmark points in the mesh;

sampling a neighborhood in each rectangular grid around the initial position for each of the connected landmark points in the mesh, the neighborhood including a number of candidate locations of the connected landmark points in the mesh;

calculating a graph data structure for all of the connected landmark points in the mesh, each node (vertex) of the graph data structure representing a connected landmark point of the connected landmark points in the mesh and each edge (arc) of the graph data structure representing the connection between a pair of the landmark points in the mesh;

calculating a path through the graph data structure wherein an order of visiting the connected landmark points in the mesh is determined by the path through the graph data structure, the path being a minimum spanning tree;

optimizing a cost function expressing a weighted sum of an overall gray level cost and an overall shape cost for all of the candidate locations of the connected landmark points in the mesh based on the path; and

calculating a segmented anatomic entity as a surface or a volume through a selected combination of the candidate locations of each of the connected landmark points in the mesh for which combination the cost function is optimized.

2. The method according to claim 1, wherein a cost associated with a candidate location represents a measure of similarity between a sampled gray level pattern at the candidate location and an expected gray level pattern at the landmark point, the measure of similarity expressed by a Mahalanobis distance defined as a distance between the sampled gray level pattern and a mean gray level pattern weighted with an inverse covariance matrix.

3. The method according to claim 2, wherein the feature image is a derivative image at a predefined scale a, or an image representation including mathematical moments of a local histogram of the digital medical image, or a derivative image in a window having a width α

around the location of a landmark point.

4. The method according to claim 2, wherein the cost associated with a candidate location is a total gray level cost obtained by summing gray level costs of the candidate location in every feature image of a set of feature images calculated for the digital medical image, the gray level costs expressed as a Mahalanobis distance defined as a distance between a gray value profile at the candidate location with a corresponding mean profile weighted by the inverse covariance matrix, the mean profile and the covariance matrix retrieved from a gray value model associated with the anatomic entity.

5. The method according to claim 4, wherein the gray value model is obtained by the steps of:

sampling a manually segmented surface or volume of the anatomic entity at a number of landmark points;

forming a polygonal mesh of the surface or a polyhedral mesh of the volume of the anatomic entity by connecting respective groups of neighboring landmark points into a polygon or a polyhedron, respectively, such that each edge of the polygon or the polyhedron is shared by at most two polygons;

defining a graph data structure for all of the landmark points, each node (vertex) representing a landmark and each edge (arc) representing a connection between a pair of landmark points;

sampling a number of points in a neighborhood of each of the landmark points;

for each landmark point, arranging sampled points in a neighborhood around a landmark point in a profile;

computing at least one feature image;

computing a mean profile for each landmark point and for each feature image;

computing a covariance matrix of the profiles for each landmark point and each feature image; and

identifying the graph, the mean profiles, and the covariance matrix as a gray value model of the anatomic entity.

6. The method according to claim 5, where the polygon is a triangle.

7. The method according to claim 5, where the polyhedron is a tetrahedron.

8. The method according to claim 1, wherein the overall gray level cost is a combination of all individual gray level costs of a profile in a number of feature images, and the individual gray level cost is a Mahalanobis distance defined as a distance between the profile and the mean profile of that feature weighted with an inverse of the covariance matrix, the mean profile and the covariance matrix being retrieved from a gray value model of the anatomic entity.

9. The method according to claim 8, wherein the gray value model is obtained by the steps of:

sampling a manually segmented surface or a volume of the anatomic entity at a number of landmark points;

forming a polygonal mesh of the surface or a polyhedral mesh of the volume of the anatomic entity by connecting respective groups of neighboring landmark points into a polygon or a polyhedron, respectively, such that each edge of the polygon or the polyhedron is shared by at most two polygons;

defining a graph data structure for all landmark points, each node (vertex) representing a landmark and each edge (arc) representing a connection between a pair of landmark points;

sampling a number of points in a neighborhood of each of the landmark points;

for each landmark point, arranging sampled points in a neighborhood around a landmark point in a profile;

computing at least one feature image;

computing a mean profile for each landmark point and for each feature image;

computing a covariance matrix of the profiles for each landmark point and each feature image; and

identifying the graph, the mean profiles, and the covariance matrix as a gray value model of the anatomic entity.

10. The method according to claim 1, wherein the overall shape cost is a combination of all individual costs of connection vectors, the connection vectors connecting successive landmark points, for all landmark points in which an individual shape cost is expressed by a Mahalanobis distance defined as a distance of a connection vector between two successive landmark points to a mean connection vector weighted with an inverse of a covariance matrix, the mean connection vector and the covariance matrix being retrieved from a shape model of the anatomic entity.

11. The method according to claim 10, wherein the shape model is obtained by the steps of:

sampling a manually segmented surface or a volume of the anatomic entity at a number of landmark points;

forming a polygonal mesh of the surface or a polyhedral mesh of the volume by connecting groups of neighboring landmark points in a polygon or a polyhedron, respectively;

defining a graph data structure for all landmark points, each node (vertex) representing a landmark point and each edge (arc) representing a connection between a pair of landmark points;

computing connection vectors between pairs of landmark points;

computing a mean connection vector between the pairs of landmark points;

computing a covariance matrix of connection vectors for each connection in the graph data structure; and

identifying the graph, the mean connection vector, and the covariance matrices as a geometric model of the anatomic entity.

12. The method according to claim 1, wherein the cost function is optimized by dynamic programming.

13. The method according to claim 1, wherein the number of candidate locations is reduced by selecting the candidate locations with a minimal total gray value cost, the total gray value cost being a sum over all feature images of all gray value costs for the candidate position in the neighborhood, the gray value cost expressed as a Mahalanobis distance defined as a distance of a gray value profile at the candidate location to a corresponding mean profile weighted with an inverse of a covariance matrix, the mean profile and the covariance matrix retrieved from a gray value model associated with the anatomic entity.

14. The method according to claim 1, wherein the gray value model and the shape model are constructed from and applied to a multi-resolution representation of the digital medical image.

15. The method according to claim 1, wherein the path through the graph data structure is obtained by breadth first traversal.

16. The method according to claim 1, further comprising the steps of:

tracking the position of the landmark points towards a final position in the digital medical image using finer resolutions of the digital medical image; and

calculating, on each resolution level of the digital medical image, a graph data structure.

17. A non-transitory computer-readable medium containing a computer program adapted to carry out, when the computer program runs on a computer, a method according to claim 1.