Image analysis

• US 7,684,596 B2
• Filed: 02/05/2004
• Issued: 03/23/2010
• Est. Priority Date: 02/11/2003
• Status: Active Grant

• Current Assignee: Qinetiq Limited (QinetiQ Group PLC)
• Original Assignee: Qinetiq Limited (QinetiQ Group PLC)
• Inventors: Watson, Graham Howard, Watson, Sharon Katrina
• Primary Examiner(s): Strege; John B
• Application Number:
  US10/543,911
• Publication Number:
  US 20060083418A1
• Time in Patent Office: 2,238 Days
• Field of Search: 382/133, 382/165, 382/199, 382/128, 382/130, 382/260, 382/261, 382/262, 340/146
• US Class Current: 382/128
• CPC Class Codes: G06T 2207/10056 Microscopic image
• CPC Class Codes: G06T 2207/30024 Cell structures in vitro; Tissue sections in vitro
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06T 7/62 of area, perimeter, diameter or volume
• CPC Class Codes: G06V 20/69 Microscopic objects, e.g. biological cells or cellular parts

Abstract

A method for the automated analysis of digital images, particularly for the purpose of assessing nuclear pleomorphism from images of histological slides for prognostication of breast cancer. The method includes the steps of identifying the locations of objects within the image which have intensity and size characteristics consistent with epithelial cell nuclei and deriving boundaries for those objects. Statistics concerning at least the shapes of the derived boundaries are calculated and clutter is rejected on the basis of those statistics and/or they are used to assign probabilities that the respective objects are epithelial cell nuclei, and a measure of the variability of at least the areas enclosed by such boundaries is then calculated.

Claims

1. A method for the automated analysis of a digital image having an array of pixels including using a computer or processor to perform the successive steps comprising:

(a) identifying locations of objects within the image which have specified intensity and size characteristics;

(b) deriving respective boundaries for respective such objects;

(c) assessing the significance of objects based on statistics concerning at least the shapes of their respective derived boundaries; and

(d) calculating a measure of the variability among a group of such objects within the image of at least the areas enclosed by their respective derived boundaries subject to the results of said assessment of significance;

wherein step (c) comprises assigning probabilities that respective objects are members of a specified class based on specified said statistics and said measure in step (d) is the standard deviation of the areas enclosed by said derived boundaries weighted in accordance with the respective probabilities.

2. A method according to claim 1 wherein step (a) comprises the application of a radially-symmetric difference filter with zero mean.

3. A method according to claim 2 wherein the image is filtered at a plurality of resolutions of increasing scale.

4. A method according to claim 2 wherein said locations are identified in accordance with the locations of respective extrema in the output of said radially-symmetric difference filter.

5. A method according to claim 1 wherein step (b) comprises:

defining regions of specified extent within the image which contain respective said objects; and

deriving from the data within respective said regions a respective closed contour (if any) consisting of points of the same specified intensity.

6. A method according to claim 5 further comprising determining whether respective said contours meet specified location and/or intensity orientation criteria and rejecting from further processing any such contours which do not.

7. A method according to claim 5 further comprising tracking respective said contours and measuring a curvature thereof.

8. A method according to claim 7 wherein said tracking and measuring of the curvature is performed by use of a Probability Density Association Filter.

11. A method according to claim 1 wherein step (c) comprises rejecting objects if the enclosed area of their respective derived boundaries is less than a specified value or if the minimum curvature of their respective derived boundaries is less than a specified value.

12. A method according to claim 1 wherein said assignment of probabilities is performed by use of a Fisher classifier.

13. A method according to claim 1 wherein the image is of a histological or cytology specimen.

14. A computer readable medium having program code recorded thereon for executing the method according to claim 1.

9. A method for the automated analysis of a digital image having an array of pixels including using a computer or processor to perform the successive steps comprising:

(a) identifying locations of objects within the image which have specified intensity and size characteristics;

(b) deriving respective boundaries for respective such objects;

(c) assessing the significance of objects based on statistics concerning at least the shapes of their respective derived boundaries; and

(d) calculating a measure of the variability among a group of such objects within the image of at least the areas enclosed by their respective derived boundaries subject to the results of said assessment of significance;

wherein step (b) comprises;

defining regions of specified extent within the image which contain respective said objects; and

deriving from the data within respective said regions a respective closed contour (if any) consisting of points of the same specified intensity

and wherein step (a) comprises the application of a radially-symmetric difference filter with zero mean and, for respective said regions, the respective said contour is derived from the output of said radially-symmetric difference filter with zero mean for the respective region.

10. A computer readable medium having program code recorded thereon for executing the method according to claim 9.

15. A method for the automated analysis of a digital image having an array of pixels including using a computer or processor to perform the successive steps comprising:

(a) identifying locations of objects within the image which have specified intensity and size characteristics;

(b) deriving respective boundaries for respective such objects;

(c) assessing the significance of objects based on statistics concerning at least the shapes of their respective derived boundaries; and

(d) calculating a measure of the variability among a group of such objects within the image of at least the areas enclosed by their respective derived boundaries subject to the results of said assessment of significance;

and further comprising, following step (a);

defining masks of specified extent within the image around the identified said locations;

calculating for respective said masks at least one intensity threshold from the pixels within the respective mask; and

thresholding the pixels within respective masks using the respective threshold(s) calculated therefor;

whereby to derive a set of binary images corresponding to said masks and containing a set of regions representing connected pixels of a specified class of intensity in relation to the applied thresholding.

16. A method according to claim 15 wherein two said thresholds are calculated for respective masks and hysteresis thresholding is applied to the pixels within respective masks using the two respective thresholds.

17. A method according to claim 15 further comprising:

calculating from data in regions of the original image corresponding to respective said binary image regions statistics concerning their size, shape and/or intensities therein; and

accepting or rejecting for further analysis objects corresponding to such regions on the basis of such statistics.

18. A method according to claim 17 comprising the step of calculating statistics from data in regions of the original image corresponding to respective said binary image regions wherein said statistics comprise:

the area of the respective region;

the thickness of the respective region (as hereinbefore defined);

the maximum intensity of the pixels within the respective region; and

the minimum intensity of the pixels within the respective region.

19. A method according to claim 17 wherein said accepting or rejecting of objects for further analysis is performed by use of a Fisher classifier.

20. A computer readable medium having program code recorded thereon for executing the method according to claim 15.

21. A method for the automated analysis of a digital image having an array of pixels including using a computer or processor to perform the successive steps comprising:

(a) identifying locations of objects within the image which have specified intensity and size characteristics;

(b) deriving respective boundaries for respective such objects;

(c) assessing the significance of objects based on statistics concerning at least the shapes of their respective derived boundaries; and

(d) calculating a measure of the variability among a group of such objects within the image of at least the areas enclosed by their respective derived boundaries subject to the results of said assessment of significance;

wherein step (b) comprises, for respective objects;

establishing a closed curve of specified form around the respective location;

deforming the curve outwardly in accordance with intensity characteristics of the image in the direction normal to the curve at a plurality of points around the curve;

repeating the aforesaid deformation until the area enclosed by the curve exceeds a specified threshold; and

selecting, from the resultant set of deformed curves, one which best meets specified characteristics of intensity gradient across the curve.

22. A method according to claim 21 wherein it is arranged that the distance by which the curve is deformed at any said point is related to the intensity gradient in the outward normal direction at that point such that the larger the gradient the smaller the deformation.

23. A method according to claim 21 wherein it is arranged that the distance by which the curve is deformed at any said point is increased if the curve is of negative curvature at that point.

24. A method according to claim 21 wherein the method for calculating the distance by which the curve is to be deformed at each such point at each such deformation includes calculating the respective intensity gradient in the outward normal direction at each such point and normalising the same.

25. A method according to claim 24 wherein said normalising comprises subtracting the respective gradient from the maximum of such gradients and dividing the result by the range in such gradients.

26. A method according to claim 24 wherein such calculating further comprises, if the curve is of negative curvature at the respective point, adding to the respective normalised gradient the absolute value of such curvature multiplied by the arc length of the curve between that point and the next point and divided by 2π

.

27. A method according to claim 26 wherein such calculating further comprises applying a smoothing filter to the results of such adding and normalizing the results of such smoothing by subtracting the minimum of such smoothed distances from each such smoothed distance, dividing the result by the range in such smoothed distances and multiplying by a prescribed number.

28. A method according to claim 21 which comprises calculating, for respective members of said set of deformed curves, a metric comprising the respective mean normal intensity gradient divided by the respective mean absolute normal intensity gradient.

29. A method according to claim 28 wherein said selecting comprises selecting a deformed curve from said set which has the highest value of said metric.

30. A method according to claim 28 wherein if none of said set of deformed curves has a value of said metric reaching a specified threshold then the respective object is rejected.

31. A method according to claim 21 which comprises calculating, for the deformed curve selected from said set, the mean value of the absolute values of the sine of the angle between the curve normal and the intensity gradient vector at a plurality of points around the curve, and rejecting the respective object if such mean value is above a specified threshold.

32. A method according claim 21 wherein step (b) comprises, for respective objects, further deforming the curve selected from the respective said set of deformed curves in accordance with intensity characteristics of the image in the direction normal to the curve at a plurality of points around the curve.

33. A method according to claim 32 wherein said further deformation comprises, at each such point, finding a position within a specified distance along the normal inward and outward of the curve where the intensity gradient is maximal.

34. A method according to claim 32 wherein said further deformation further comprises at each such point, deforming the curve to the position of the first maximum in intensity in the normal direction outwardly of said position of maximal intensity gradient within a specified distance, or if there is no such maximum, deforming the curve to the nearest position in the normal direction outwardly of said position of maximal intensity gradient within a specified distance where the intensity is within a specified proportion of the intensity at said distance.

35. A computer readable medium having program code recorded thereon for executing the method according to claim 21.

36. A method for the automated analysis of a digital image having an array of pixels including using a computer or processor to perform the successive steps comprising:

(a) identifying locations of objects within the image which have specified intensity and size characteristics;

(b) deriving respective boundaries for respective such objects;

(c) assessing the significance of objects based on statistics concerning at least the shapes of their respective derived boundaries; and

(d) calculating a measure of the variability among a group of such objects within the image of at least the areas enclosed by their respective derived boundaries subject to the results of said assessment of significance;

wherein step (c) comprises assigning probabilities that respective objects are members of a specified class based on statistics comprising;

the area enclosed by the derived boundary;

the perimeter of the derived boundary squared divided by the area enclosed by the derived boundary; and

the minimum curvature of the derived boundary.

37. A computer readable medium having program code recorded thereon for executing the method according to claim 36.

38. A method for the automated analysis of a digital image having an array of pixels including using a computer or processor to perform the successive steps comprising:

(a) identifying locations of objects within the image which have specified intensity and size characteristics;

(b) deriving respective boundaries for respective such objects;

(c) assessing the significance of objects based on statistics concerning at least the shapes of their respective derived boundaries; and

(d) calculating a measure of the variability among a group of such objects within the image of at least the areas enclosed by their respective derived boundaries subject to the results of said assessment of significance;

wherein step (c) comprises assigning probabilities that respective objects are members of a specified class based on statistics comprising;

the area enclosed by the derived boundary;

the perimeter of the derived boundary squared divided by the area enclosed by the derived boundary;

the maximum intensity of the original image pixels within the derived boundary; and

the minimum intensity of the original image pixels within the derived boundary.

39. A computer readable medium having program code recorded thereon for executing the method according to claim 38.

40. A method for the automated analysis of a digital image having an array of pixels including using a computer or processor to perform the successive steps comprising:

(a) identifying locations of objects within the image which have specified intensity and size characteristics;

(b) deriving respective boundaries for respective such objects;

(c) assessing the significance of objects based on statistics concerning at least the shapes of their respective derived boundaries; and

(d) calculating a measure of the variability among a group of such objects within the image of at least the areas enclosed by their respective derived boundaries subject to the results of said assessment of significance;

wherein step (c) comprises assigning probabilities that respective objects are members of a specified class based on specified said statistics, the image is of a section of breast tissue and said specified class is identified as the class of epithelial cell nuclei.

41. A computer readable medium having program code recorded thereon for executing the method according to claim 40.