Method for Segmenting Digital Medical Image

• US 20080101678A1
• Filed: 10/24/2007
• Published: 05/01/2008
• Est. Priority Date: 10/25/2006
• Status: Active Grant

• Current Assignee: Agfa Nv (Agfa-Gevaert N.V.)
• Original Assignee: AGFA Healthcare NV (Agfa-Gevaert N.V.)
• Inventors: Suliga, Marek, Nyssen, Edgard, Dewaele, Piet, Behiels, Gert, Deklerck, Rudi
• Granted Patent: US 8,290,223 B2
• Time in Patent Office: Days
• US Class Current: 382/132
• CPC Class Codes: G06T 2207/10116 X-ray image
• CPC Class Codes: G06T 2207/30068 Mammography; Breast
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling
• CPC Class Codes: G06V 10/426 Graphical representations
• CPC Class Codes: G06V 10/462 Salient features, e.g. scale invariant feature transforms [SIFT]

Abstract

A Markov Random Field (MRF)-based technique is described for performing clustering of images characterized by poor or limited data. The proposed method is a statistical classification model that labels the image pixels based on the description of their statistical and contextual information. Apart from evaluating the pixel statistics that originate from the definition of the K-means clustering scheme, the model expands the analysis by the description of the spatial dependence between pixels and their labels (context), hence leading to the reduction of the inhomogeneity of the segmentation output with respect to the result of pure K-means clustering.

Claims

1. A method for segmenting a digital medical image comprising the steps of

a. computing an initial segmentation class label for each pixel on the basis of a K-means clustering algorithm and computing associated class centroids and gamma functions, a gamma, function weighting the importance of likelihood potential and clique potential in an energy function;

b. computing an energy measure for a current pixel associated with assigning each label in turn to the pixel, the energy measure being a sum of two components;

a measure of the distance between the pixel'"'"'s feature value and the value of the centroid indicated by the current segmentation label, and

a measure of the difference of the assigned label and the label of at least one neighboring pixel;

c. selecting the label for the current pixel yielding the lowest energy;

d. recalculating the segmentation class centroids and gamma functions;

e. proceeding to the next pixel until all pixels in the image are visited;

f. iterating steps b-e until the class labels of the image pixels do not change so as to yield the final segmentation of the image.

2. A method according to claim 1 wherein the measure of the distance is the Euclidian distance.

3. A method according to claim 1 wherein the measure of the distance is a weighted distance.

4. A method according to claim 1 wherein the measure of the distance is the Mahalanobis distance.

5. A method according to claim 1 wherein the neighboring pixels belong to an isotropic neighborhood.

6. A method according to claim 5 wherein the neighborhood is confined to a 3×

3 neighborhood.

7. A method according to claim 1 wherein neighboring pixels are selected according to a specific local orientation.

8. A method according to claim 1 wherein a pixel is represented by its grey value.

9. A method according to claim 1 wherein a pixel is represented by a multi-dimensional feature provided by a multi-dimensional texture operator.

10. A method according to claim 9 wherein the multi-dimensional feature is calculated on the basis of a filter bank of local derivatives.

11. A method according to claim 1 wherein the sequence of steps (d) and (e) is interchanged.

17. A method according to claim 1 applied to a mammographic image.

12. A method, for segmenting a digital medical image comprising the steps of:

a) computing a multi-resolution (MR) representation of the digital image;

b) selecting an image at a coarse resolution level in the MR representation as initial image and applying a K-means clustering algorithm to the initial image,

c) computing an energy measure for a current pixel associated with assigning each label in turn to the pixel, the energy measure being a sum of two components;

a measure of the distance between the pixel'"'"'s feature value and the value of the centroid indicated by the current segmentation label, and

a measure of the difference of the assigned label and the label of at least one neighboring pixel;

d) selecting the label for the current pixel yielding the lowest energy;

e) recalculating the segmentation class centroids and gamma functions;

f) proceeding to the next pixel until all pixels in the image are visited;

g) iterating steps b to f until the class labels of the image pixels do not change so as to yield the final segmentation of the image;

j) propagating the labels to the next finer resolution level of the multi-resolution representation;

h) iterating step (c) to (j) until the original image is segmented.

13. A method according to claim 12 wherein the multi-resolution representation is a Gaussian pyramid representation.

14. A method according to claim 12 wherein the multi-resolution representation is a quadtree.

15. A method according to claim 12 wherein propagating is performed by locally replicating the labels to an image at a higher resolution level in the multi-resolution representation.

16. A method according to claim 12 wherein the sequence of steps (e) and (f) is interchanged.

18. A method according to claim 12 applied to a mammographic image.

19. A computer software product for segmenting a digital medical image, the product comprising a computer-readable medium in which program instructions are stored, which instructions, when read by a computer, cause the computer to:

a. compute an initial segmentation class label for each pixel on the basis of a K-means clustering algorithm and computing associated class centroids and gamma functions a gamma function weighting the importance of likelihood potential and clique potential in an energy function;

b. compute an energy measure for a current pixel associated with assigning each label in turn to the pixel, the energy measure being a sum of two components;

a measure of the distance between the pixel'"'"'s feature value and the value of the centroid indicated by the current segmentation label, and

a measure of the difference of the assigned label and the label of at least one neighboring pixel;

c. select the label for the current pixel yielding the lowest energy;

d. recalculate the segmentation class centroids and gamma functions;

e. proceed to the next pixel until all pixels in the image are visited;

f. iterate steps b-e until the class labels of the image pixels do not change so as to yield the final segmentation of the image.

20. A computer software product for segmenting a digital medical image, the product comprising a computer-readable medium in which program instructions are stored, which instructions, when read by a computer, cause the computer to:

a) compute a multi-resolution (MR) representation of the digital image;

b) select an image at a coarse resolution level in the MR representation as initial image and applying a K-means clustering, algorithm to the initial image,

c) compute an energy measure for a current pixel associated with assigning each label in turn to the pixel, the energy measure being a sum of two components;

a measure of the distance between the pixel'"'"'s feature value and the value of the centroid indicated by the current segmentation label, and

a measure of the difference of the assigned label and the label of at least one neighboring pixel;

d) select the label for the current pixel yielding the lowest energy;

e) recalculate the segmentation class centroids and gamma functions;

f) proceed to the next pixel until all pixels in the image are visited;

g) iterate steps b to f until the class labels of the image pixels do not change so as to yield the final segmentation of the image;

j) propagate the labels to the next finer resolution level of the multi-resolution representation;

h) iterate step (c) to (i) until the original image is segmented.