Method and system for using local attention in the detection of abnormalities in digitized medical images

• US 6,014,452 A
• Filed: 07/28/1997
• Issued: 01/11/2000
• Est. Priority Date: 07/28/1997
• Status: Expired due to Term

• Current Assignee: Hologic Incorporated
• Original Assignee: R2 Technology, Inc. (Hologic Incorporated)
• Inventors: Roehrig, Jimmy R., Zhang, Wei, Romsdahl, Harlan M.
• Primary Examiner(s): Chang, Jon
• Assistant Examiner(s): Dastouri, Mehrdad
• Application Number:
  US08/901,541
• Time in Patent Office: 897 Days
• Field of Search: 382/130, 382/131, 382/132, 382/156, 382/225, 378/37
• US Class Current: 382/132
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06V 10/267 by performing operations on regions, e.g. growing, shrinking or watersheds
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images

Abstract

A method an system for using a local attention threshold to aid in the detection of clustered abnormalities in digitized medical images is disclosed. The local attention threshold is applied to locate spots within a predetermined distance from previously identified spots. More specifically, seed pixels are identified by applying a first seed threshold function to the output of a shift-invariant neural network and adaptive threshold. The seed pixels are then segmented into spots by applying a segmentation threshold function to each seed pixel. False-positive spots are removed using various techniques. Additional seed pixels are then identified by applying a local attention threshold to pixels within a predetermined distance to previously identified spots. The local attention threshold disclosed is less selective for pixels which are closer to the nearest spot than for pixels which are further from the nearest spot. The new seed pixels are then segmented into spots, and potential abnormalities are identified in the medical image based in part on the closeness of the identified spots.

Claims

1. A method for detecting clustered abnormalities in a digitized medical image comprising the steps of:

receiving digital image data, said image data representing a medical image and comprising a plurality of pixels, each pixel having a gray-scale value based at least in part on the brightness of the pixel;

obtaining an output value for at least some of the pixels, the output value of a pixel being based at least in part on the contrast of the gray-scale value of the pixel relative to the gray-scale values of nearby pixels;

identifying a first set of seed pixels by applying a first threshold function to the output values of at least some of the pixels;

identifying a first set of groups of pixels by applying a second threshold function to the output values of a plurality of pixels in close proximity to at least one pixel of the first set of seed pixels;

identifying a second set of seed pixels by applying a third threshold function to the output values of a plurality of pixels within a predetermined distance to at least some pixels of the first set of groups, said third threshold function being less selective than the first threshold function;

identifying a second set of groups of pixels by applying a fourth threshold function to the output values of a plurality of pixels in close proximity to at least one pixel of the second set of seed pixels; and

identifying potential abnormalities in the medical image according to at least the distance between at least some of the groups in the first and second sets of groups of pixels.

2. The method of claim 1 wherein the third threshold function is less selective for pixels which are closer to the nearest group of pixels in the first set of groups than for pixels which are further from the nearest group of pixels in the first set of groups.

3. The method of claim 2 wherein the medical image is an x-ray mammogram and the clustered abnormalities are clustered microcalcifications.

4. The method of claim 3 wherein the predetermined distance corresponds to a distance between about 1 and 5 millimeters on the x-ray mammogram.

5. The method of claim 4 wherein the predetermined distance corresponds to a distance between about 2 and 3 millimeters on the x-ray mammogram.

6. The method of claim 5 wherein the predetermined distance corresponds to a distance of about 2.5 millimeters on the x-ray mammogram.

7. The method of claim 1 wherein the first and second threshold functions are fixed constants.

8. The method of claim 7 wherein the second and fourth threshold functions are of equivalent selectivity.

9. The method of claim 8 wherein each group of pixels in the first and second sets of groups of pixels consists of groups of two or more pixels in which each pixel is horizontally, vertically or diagonally adjacent with at least one other pixel in the group.

10. The method of claim 1 wherein the output value is obtained by inputting the gray-scale values for a plurality of pixels into a feed-forward shift-invariant neural network.

11. The method of claim 1 further comprising the steps of:

analyzing features of each of the groups of pixels in the first set of groups to identify groups which are not likely to indicate abnormalities in the medical image; and

removing from the first set of groups the groups identified as not likely to indicate abnormalities, said steps of analyzing and removing being performed prior to the step of identifying a second set of seed pixels.

12. The method of claim 1 further comprising the step of applying a fixed threshold to the output values of a plurality of pixels to select a sub-image set of pixels, the sub-image set of pixels being a small fraction of the total number of pixels in the digitized medical image, wherein the step of identifying the first set of seed pixels comprises applying the first threshold function to the output value of each pixel in the sub-image set of pixels.

13. A method for detecting clustered abnormalities in a digitized medical image comprising the steps of:

receiving digital image data, said image data representing a medical image and comprising a plurality of pixels, each pixel having a gray-scale value based at least in part on the brightness of the pixel;

obtaining an output value for at least some of the pixels, the output value of a pixel being obtained by inputting the gray-scale values for a plurality of pixels into a feed-forward shift-invariant neural network, the output value being in part based on the contrast of the gray-scale value of the pixel relative to the gray-scale value of nearby pixels;

applying a fixed threshold to the output values of a plurality of pixels to select a set of candidate pixels;

identifying a first set of seed pixels by applying a first threshold function to the output values of each of the candidate pixels;

identifying a first set of groups of pixels by applying a second threshold function to a plurality of pixels in close proximity to at least one pixel of the first set of seed pixels;

analyzing features of each of the groups of pixels in the first set of groups to identify groups which are not likely to indicate malignant tumors in the medical image;

removing from the first set of groups the groups identified as not likely to indicate malignant tumors;

identifying a second set of seed pixels by applying a third threshold function to the output values of pixels in the set of candidate pixels that are within a predetermined distance to at least one group of the first set of groups, said third threshold function being less selective than the first threshold function, and being less selective for pixels which are closer to the nearest group of pixels in the first set of groups than for pixels which are further from the nearest group of pixels in the first set of groups;

identifying a second set of groups of pixels by applying the second threshold function to the output values of each of a plurality of pixels in close proximity to at least one pixel of the second set of seed pixels; and

identifying potential abnormalities in the medical image according to at least the distance between at least some of the groups in the first and second set of groups of pixels.

14. A system for detecting clustered abnormalities in digitized medical images, said system comprising:

a source of digitized image data, said image data representing a medical image and comprising a plurality of pixels, each pixel having a gray-scale value based at least in part on the brightness of the pixel;

an image processor adapted to produce an output value for at least some of the pixels, the output value of a pixel being based at least in part on the contrast of the gray-scale value of the pixel relative to the gray-scale value of nearby pixels;

a first selector adapted to select a first set of seed pixels by applying a first threshold function to the output values of at least some of the pixels;

a first segmenter adapted to select a first set of groups of pixels by applying a second threshold function to the output values of a plurality of pixels in close proximity to at least one pixel of the first set of seed pixels;

a second selector adapted to select a second set of seed pixels by applying a third threshold function to the output values of a plurality of pixels within a predetermined distance to at least some pixels of the first set of groups, said third threshold function being less selective than the first threshold function;

a second segmenter adapted to select a second set of groups of pixels by applying a fourth threshold function to the output values of a plurality of pixels in close proximity to at least one pixel of the second set of seed pixels; and

an indicator for indicating potential abnormalities in the medical image according at least the distance between at least some of the groups in the first and second set of groups of pixels.

15. The system of claim 14 wherein the third threshold function is less selective for pixels which are closer to the nearest group of pixels in the first set of groups than for pixels which are further from the nearest group of pixels in the first set of groups.

16. The system of claim 14 wherein the medical image is an x-ray mammogram, the clustered abnormalities are clustered microcalcifications, and the predetermined distance corresponds to a distance between about 2 and 3 millimeters on the x-ray mammogram.

17. The system of claim 14 wherein the first, second and fourth threshold functions are fixed constants, the second and fourth threshold functions are of equivalent selectivity, and each group of pixels in the first and second sets of groups of pixels consists of groups of two or more pixels in which each pixel is horizontally, vertically or diagonally adjacent with at least one other pixel in the group.

18. The system of claim 14 wherein the output value is obtained by inputting the gray-scale values for a plurality of pixels into a feed-forward shift-invariant neural network.

19. The system of claim 18 further comprising:

a sub-image selector adapted to apply a fixed threshold to the output values of a plurality of pixels to select a sub-image set of pixels, the sub-image set of pixels being a small fraction of the total number of pixels in the digitized medical image, wherein first selector applies the first threshold function to the output value of each pixel in the sub-image set of pixels; and

a deselector adapted to analyze features of each of the groups of pixels in the first set of groups to identify groups which are not likely to indicate abnormalities in the medical image, and remove from the first set of groups the groups identified as not likely to indicate abnormalities.

20. A computer-readable medium which can be used for directing an apparatus to detect clustered microcalcifications in an image comprising:

means for receiving digital image data, said image data representing a medical image and comprising a plurality of pixels, each pixel having a gray-scale value based at least in part on the brightness of the pixel;

means for obtaining an output value for at least some of the pixels, the output value of a pixel being based at least in part on the contrast of the gray-scale value of the pixel relative to the gray-scale values of nearby pixels;

means for identifying a first set of seed pixels by applying a first threshold function to the output values of at least some of the pixels;

means for identifying a first set of groups of pixels by applying a second threshold function to the output values of a plurality of pixels in close proximity to at least one pixel of the first set of seed pixels;

means for identifying a second set of seed pixels by applying a third threshold function to the output values of a plurality of pixels within a predetermined distance to at least some pixels of the first set of groups, said third threshold function being less selective than the first threshold function;

means for identifying a second set of groups of pixels by applying a fourth threshold function to the output values of a plurality of pixels in close proximity to at least one pixel of the second set of seed pixels; and

means for identifying potential abnormalities in the medical image according to at least the distance between at least some of the groups in the first and second sets of groups of pixels.

21. The computer-readable medium of claim 20 wherein the first, second and fourth threshold functions are fixed constants, the second and fourth threshold functions are of equivalent selectivity, and the third threshold function is less selective for pixels which are closer to the nearest group of pixels in the first set of groups than for pixels which are further from the nearest group of pixels in the first set of groups.

22. The computer-readable medium of claim 21 wherein the output value is obtained by inputting the gray-scale values for a plurality of pixels into a feed-forward shift-invariant neural network, the medical image is an x-ray mammogram and the clustered abnormalities are clustered microcalcifications, and the predetermined distance corresponds to a distance between about 1 and 5 millimeters on the x-ray mammogram.

23. The computer-readable medium of claim 22 wherein each group of pixels in the first and second sets of groups of pixels consists of groups of two or more pixels in which each pixel is horizontally, vertically or diagonally adjacent with at least one other pixel in the group.

24. The computer-readable medium of claim 23 further comprising:

means for analyzing features of each of the groups of pixels in the first set of groups to identify groups which are not likely to indicate abnormalities in the medical image; and

means for removing from the first set of groups the groups identified as not likely to indicate abnormalities.

25. The computer-readable medium of claim 24 further comprising means for applying a fixed threshold to the output values of a plurality of pixels to select a sub-image set of pixels, the sub-image set of pixels being a small fraction of the total number of pixels in the digitized medical image, wherein the means for identifying the first set of seed pixels comprises means for applying the first threshold function to the output value of each pixel in the sub-image set of pixels.