Granted Patent B2

US 8,218,850 · App. 12/317,530 · Granted Jul 10, 2012

Abstract

Claims (32)

1. A method of processing a mammogram image to derive a value for a parameter useful in detecting differences in breast tissue in subsequent images of the same breast or relative to a control group of such images, said derived parameter being an aggregate probability score reflecting the probability of the image being a member of a predefined class of mammogram images, said method comprising computing for each of a multitude of pixels within a large region of interest within the image a pixel probability score assigned by a trained statistical classifier according to the probability of said pixel belonging to an image belonging to said class, said pixel probability being calculated on the basis of a selected plurality of features of said pixels, and computing said parameter by aggregating the pixel probability scores over said region of interest.

2. A method as claimed in claim 1 , wherein said trained classifier is trained by unsupervised learning.

3. A method as claimed in claim 1 , wherein said trained classifier is trained by supervised learning.

4. A method as claimed in claim 1 , wherein said classifier is trained on a set of images associated with a higher breast density and a set of images associated with a lower breast density.

5. A method as claimed in claim 1 , wherein said classifier is trained on a set of images associated with a higher risk of breast cancer and a set of images associated with a lower risk of breast cancer.

6. A method according to claim 1 , wherein a said pixel feature on the basis of which each pixel is classified is a quotient value representative of the aspect ratio of a tissue structure depicted in the image in the to which structure said pixel belongs.

7. A method as claimed in claim 6 , wherein said quotient value is determined for each said pixel at each of a plurality of scales.

8. A method as claimed in claim 7 , wherein said quotient values are determined as the normalised difference between eigenvalues of a Hessian matrix based on Gaussian derivatives at a predetermined scale of pixels of the image, which Gaussian derivatives relate the intensity of each pixel to the intensities of the neighbours of said pixel.

9. A method as claimed in claim 1 , wherein a said pixel feature on the basis of which each pixel is classified is a selected derivative from the set of n local, partial derivatives up to the order n (n-jet).

10. A method as claimed in claim 9 , wherein the n-jet is implemented as Gaussian derivatives at a plurality of scales.

11. A method as claimed in claim 9 , wherein a further feature on the basis of which each pixel is classified is a measure of its location in said mammogram image.

12. A method as claimed in claim 9 , wherein said n-jet is the 3-jet.

13. A method as claimed in claim 12 , wherein said features include the third order horizontal derivatives in the 3-jet set.

14. A method as claimed in claim 1 or claim 2 , comprising

plotting values for said features of each pixel on a pre-determined model of a cluster map, said cluster map having at least two pre-plotted points;

clustering the plotted values around said at least two pre-plotted points to result in at least two discrete clusters; and

deriving said aggregated pixel probability score by computing the areas of said clusters.

15. A method as claimed in claim 14 , wherein the pre-determined model is defined in 3-dimensional space in which the three dimensions respectively represent a quotient value determined as the normalised difference between eigenvalues of a Hessian matrix based on Gaussian derivatives at a predetermined scale of pixels of the image, which Gaussian derivatives relate the intensity of each pixel to the intensities of the neighbours of said pixel, when calculated at a respective one of three different scales.

16. A method as claimed in claim 14 , comprising a preliminary step of preparing said pre-determined model of the cluster map incorporating said at least two pre-plotted points.

17. A method as claimed in claim 14 , wherein the step of preparing the model of the cluster map comprises:

a) computing a quotient value for pixels in a test image of a breast based on the results of Hessian matrices compiled for said pixels in the image;

b) plotting each quotient value in 3-dimensional space;

c) randomly selecting at least two points amongst the quotient values as starting points;

d) calculating the distance between each quotient value and each starting point and assigning each quotient value to the starting point to which it is nearest;

e) calculating a mean of each group of points and redefining the starting point as the calculated mean point;

iteratively repeating steps d) and e) until the calculated mean point is equal to the starting point, whereby the final calculated mean points are arranged to be the pre-plotted points of the model of the cluster map.

18. A method as claimed in claim 14 , wherein the step of clustering further comprises:

calculating which of said at least two pre-plotted points each plotted value is closest to; and

assigning each value accordingly to said closest pre-plotted point to result in said two discrete clusters of plotted points.

19. A method as claimed in claims 16 , wherein the number of pre-plotted points determines the number of resulting clusters.

20. A computer programmed to operate a method as claimed in claim 1 .

21. An instruction set for directing a computer to operate a method as claimed in claim 1 .

Assignments (24)