US5491627A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Method and system for the detection of microcalcifications in digital mammograms
Assignee: Arch Development Corp
•
Filing: 1993-05-13
•
Granted: 1996-02-13
•
Expires: 2013-05-13
Abstract
Abstract A method and system for the detection of microcalcifications in digital mammograms. Digital mammograms are obtained and regions-of-interest (ROIs) are selected therefrom which contain suspected microcalcifications, either individual or clustered microcalcifications. The suspect ROIs are background-trend corrected, followed by Fourier transformation and power spectrum calculation to perform detection in the frequency domain. Detection can also be carried out in the spatial domain by omitting the Fourier transformation and power spectrum calculation. The ROI is then scaled for input into a neural network trained to detect microcalcifications. The neural network outputs ROIs with detected microcalcifications. The method and system can also include normalizing the background-trend corrected ROIs and imputing the normalized ROI to a shift-invariant neural network trained to detect microcalcifications. The output ROI of the shift-invariant neural network is thresholded to remove additional false positive detections, and then the thresholded ROI undergoes a cluster detection to detect clustered microcalcifications. Feature extraction techniques can be applied to the remaining ROI to remove additional false positive detections.
Inventors
- 👤 Wei Zhang
- 👤 Kunio Doi
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-11-02 18:27
CPC Classifications
G; G06; G06T; G06T7/00; G06T7/0002; G06T7/0012; G16; G16H; G16H50/00; G16H50/20; G06T2207/00; G06T2207/10; G06T2207/10116; G06T2207/20; G06T2207/20084; G06T2207/30; G06T2207/30004; G06T2207/30068; G16H30/00; G16H30/20; Y; Y10; Y10S; Y10S128/00; Y10S128/92; Y10S128/923; Y10S128/924; Y10S128/925
Claims Analysis
Total Claims: 53
Claim 1
Independent Claim
77 words
A method of detecting microcalcifications in a digital mammogram, comprising: obtaining a digital mammogram; extracting regions of interest from said digital mammogram suspected of containing a microcalcification; converting said regions of interest into corresponding numerical data; inputting said numerical data to a shift-invariant neural network trained to detect microcalcifications for processing of said numerical data by said neural network, said neural network outputting corresponding output images; detecting a microcalcification in said digital mammogram using said output images.
Claim 2
Independent Claim
77 words
A method of detecting microcalcifications in a digital mammogram, comprising: obtaining a digital mammogram; extracting regions of interest from said digital mammogram suspected of containing a microcalcification; converting said regions of interest into corresponding numerical data; inputting said numerical data to a shift-invariant neural network trained to detect microcalcifications for processing of said numerical data by said neural network, said neural network outputting corresponding output images; detecting a microcalcification in said digital mammogram using said output images.
Claim 3
Dependent Claim
23 words
A method as recited in claim 1, wherein extracting regions of interest comprises extracting regions of interest suspected of containing a clustered microcalcification.
Claim 4
Dependent Claim
47 words
A method as recited in claim 2, wherein said extracting of regions of interest suspected of containing clustered microcalcifications includes selecting regions of interest containing false-positive microcalcifications; and wherein processing by said neural network includes removing a portion of said false-positive microcalcifications selected in said extracting step.
Claim 5
Dependent Claim
25 words
A method as recited in claim 3, wherein extracting a region of interest comprises extracting a region of interest suspected of containing an individual microcalcification.
Claim 6
Dependent Claim
38 words
A method as recited in claim 1, wherein processing said region of interest comprises: background-trend correcting said regions of interest to produce background-trend corrected regions of interest; and converting said background-trend corrected region of interest into numerical data.
Claim 7
Dependent Claim
46 words
A method as recited in claim 1, wherein said extracting regions of interest suspected of containing a microcalcification includes selecting regions of interest containing false-positive microcalcifications; and wherein processing by said neural network includes removing a portion of said false-positive microcalcifications selected in said extracting step.
Claim 8
Dependent Claim
20 words
A method as recited in claim 1, wherein extracting said regions of interest comprises extracting spatial domain regions of interest.
Claim 9
Independent Claim
123 words
A system for detecting microcalcifications in a digital mammogram, comprising: a device for obtaining a digital mammogram; a detector connected to said device which detects suspected microcalcifications in said digital mammogram; a region of interest selector, connected to said device and said detector, which selects regions of interest in said digital mammogram corresponding to said suspected microcalcifications; a region of interest processing device outputting processed regions of interest; an input data scaler connected to said region of interest processing device for numerically scaling said processed regions of interest; a shift-invariant neural network receiving as input data said numerically scaled regions of interest from said input data scaler and outputting corresponding output images; and a microcalcification detector for detecting microcalcifications in said output images.
Claim 10
Independent Claim
123 words
A system for detecting microcalcifications in a digital mammogram, comprising: a device for obtaining a digital mammogram; a detector connected to said device which detects suspected microcalcifications in said digital mammogram; a region of interest selector, connected to said device and said detector, which selects regions of interest in said digital mammogram corresponding to said suspected microcalcifications; a region of interest processing device outputting processed regions of interest; an input data scaler connected to said region of interest processing device for numerically scaling said processed regions of interest; a shift-invariant neural network receiving as input data said numerically scaled regions of interest from said input data scaler and outputting corresponding output images; and a microcalcification detector for detecting microcalcifications in said output images.
Claim 11
Dependent Claim
21 words
A system as recited in claim 8, wherein said region of interest extractor comprises a spatial domain region of interest extractor.
Claim 12
Independent Claim
78 words
A method of detecting clustered microcalcifications in a digital mammogram, comprising: obtaining a digital mammogram; extracting regions of interest from said digital mammogram suspected of containing a clustered microcalcification; converting said regions of interest into corresponding numerical data; inputting said numerical data into a shift-invariant neural network trained to detect clustered microcalcifications; processing said numerical data by said shift-invariant neural network to produce corresponding output images; detecting a clustered microcalcification in said digital mammogram using said output images.
Claim 13
Independent Claim
78 words
A method of detecting clustered microcalcifications in a digital mammogram, comprising: obtaining a digital mammogram; extracting regions of interest from said digital mammogram suspected of containing a clustered microcalcification; converting said regions of interest into corresponding numerical data; inputting said numerical data into a shift-invariant neural network trained to detect clustered microcalcifications; processing said numerical data by said shift-invariant neural network to produce corresponding output images; detecting a clustered microcalcification in said digital mammogram using said output images.
Claim 14
Dependent Claim
52 words
A method as recited in claim 10, wherein said extracting regions of interest suspected of containing a microcalcification includes selecting regions of interest containing false-positive microcalcifications using a computer-aided diagnostic technique; and wherein processing by said shift-invariant neural network includes removing a portion of said false-positive microcalcifications selected in said extracting step.
Claim 15
Dependent Claim
41 words
A method as recited in claim 10, wherein extracting said regions of interest comprises: detecting suspected clustered microcalcifications in said digital mammogram using a computer-aided diagnostic technique; selecting regions of interest from said digital mammogram corresponding to said suspected clustered microcalcifications.
Claim 16
Dependent Claim
93 words
A method as recited in claim 10, further comprising: processing said input data using a neural network having an input layer having a plurality of input units, at least one hidden layer having a plurality of groups, each of said groups having a plurality of units, an output layer, and connection weights between said layers; and selecting said connection weights to be shift-invariant as w(i,j;x,y)=w(i-x,j-y) where w(i,j;x,y) denotes a connection weight between a unit at a location (i,j) in a first layer and a unit at a location (x,y) in a subsequent layer.
Claim 17
Dependent Claim
165 words
A method as recited in claim 13, further comprising: training said neural network using a modified error back propagation method: ##EQU7## and in the case of said output layer, δ.sub.l.sup.L (x,y)=ƒ'(N.sub.l.sup.L (x,y)+b.sub.l.sup.L)[T(x,y)-O.sub.l.sup.L (x,y)] where: N p l (x,y) is an input of the unit at said location (x,y); W p ,q l (x, y) is the connection weights; O p l (x,y) is an output of the unit at said location (x,y); l=(1,2, . . . L) is a layer number; p=(1,2, . . . P l ) is a group number in a lth layer; q=(1,2, . . . P l +1) is a group number in a (l+1)th layer; ƒ(x) is a thresholding function; ƒ'(x) is a derivative of ƒ(x); Δw p ,q l (x,y) is a change of the connection weight; Δb p l is a change in a bias; η is a learning rate; α is a the constant factor of a moment term; and T(x,y) is a desired output image.
Claim 18
Dependent Claim
42 words
A method as recited in claim 13, further comprising: each unit in a group receiving input from a predetermined number of units in a preceding layer; and selecting said predetermined number based upon a number of individual microcalcifications in a clustered microcalcification.
Claim 19
Dependent Claim
70 words
A method as recited in claim 13, further comprising: processing said input data using a neural network having one input layer having one of said groups, a first hidden layer having two of said groups, a second hidden layer having one of said groups, and one output layer having one of said groups; and each of said units receiving input from a predetermined number of units in a preceding layer.
Claim 20
Dependent Claim
50 words
A method as recited in claim 10, wherein said converting comprises: correcting a background-trend of said regions of interest to produce corresponding background-trend corrected regions of interest; normalizing said background-trend corrected regions of interest to produce corresponding normalized regions of interest; scaling said normalized regions of interest into numerical data.
Claim 21
Dependent Claim
76 words
A method as recited in claim 17, wherein said correcting comprises: fitting a predetermined curve to said regions of interest to produce corresponding estimated background trends; and subtracting said estimated background trends from corresponding of said regions of interest; and wherein said normalizing comprises: ##EQU8## where: i l (x,y) and i(x,y) denote said regions of interest after background-trend correction and after normalization, respectively; and N 2 is a number of pixels in said regions of interest.
Claim 22
Dependent Claim
55 words
A method as recited in claim 10, further comprising: said shift-invariant neural network outputting said output images having areas corresponding to detected microcalcifications; counting said areas for each of said output images; determining a region of interest to be a positive region of interest if a number of areas is at least a predetermined value.
Claim 23
Dependent Claim
30 words
A method as recited in claim 19, further comprising: determining a region of interest to be a positive region of interest if said number of areas is at least two.
Claim 24
Dependent Claim
41 words
A method as recited in claim 10, wherein said detecting comprises: image thresholding said output images to yield an output image at a predetermined threshold level of an output image intensity value; and detecting microcalcifications using said image thresholded output images.
Claim 25
Dependent Claim
24 words
A method as recited in claim 10, further comprising: performing image feature analysis on said regions of interest suspected of containing a clustered microcalcification.
Claim 26
Dependent Claim
50 words
A method as recited in claim 10, further comprising: selecting second regions of interest in said digital mammogram corresponding to said clustered microcalcifications detected in said output images; extracting image features from said second regions of interest; performing image feature analysis on said extracted image features to identify clustered microcalcifications.
Claim 27
Dependent Claim
15 words
A method as recited in claim 23, wherein performing image feature analysis comprises feature thresholding.
Claim 28
Dependent Claim
43 words
A method as recited in claim 23, further comprising: comparing clustered microcalcifications detected by both said shift-invariant neural network and said image feature analysis with true clustered microcalcifications; and removing false-positive microcalcifications from said second regions of interest using said image feature analysis.
Claim 29
Dependent Claim
18 words
A method as recited in claim 26, wherein removing said false-positives comprises: feature thresholding said image feature analysis.
Claim 30
Dependent Claim
102 words
A method as recited in claim 26, wherein feature thresholding comprises performing at least one of: determining a contrast of said detected microcalcification; determining a size of said detected microcalcification; determining a first moment of a power spectrum of said second regions of interest; determining edge gradients in said second regions of interest; determining cumulative edge gradients in said second regions of interest; determining an average of an edge-gradient-orientation histogram of said second regions of interest; determining a standard deviation of an edge-gradient-orientation histogram of said second regions of interest; and determining an average optical density of said second regions of interest.
Claim 31
Dependent Claim
20 words
A method as recited in claim 10, wherein extracting said regions of interest comprises extracting spatial domain regions of interest.
Claim 32
Independent Claim
128 words
A system for detecting clustered microcalcifications in a digital mammogram, comprising: a device for obtaining a digital mammogram; a detector connected to said device which detects suspected microcalcifications in said digital mammogram; a region of interest selector, connected to said device and said detector, which selects regions of interest in said digital mammogram corresponding to said suspected microcalcifications; a region of interest processing device outputting processed regions of interest; an input data scaler connected to said region of interest processing device for numerically scaling said processed regions of interest; a shift-invariant neural network trained to detect clustered microcalcifications receiving as input data said numerically scaled regions of interest from said data scaler and outputting corresponding output images; and a microcalcification detector for detecting microcalcifications in said output images.
Claim 33
Independent Claim
128 words
A system for detecting clustered microcalcifications in a digital mammogram, comprising: a device for obtaining a digital mammogram; a detector connected to said device which detects suspected microcalcifications in said digital mammogram; a region of interest selector, connected to said device and said detector, which selects regions of interest in said digital mammogram corresponding to said suspected microcalcifications; a region of interest processing device outputting processed regions of interest; an input data scaler connected to said region of interest processing device for numerically scaling said processed regions of interest; a shift-invariant neural network trained to detect clustered microcalcifications receiving as input data said numerically scaled regions of interest from said data scaler and outputting corresponding output images; and a microcalcification detector for detecting microcalcifications in said output images.
Claim 34
Dependent Claim
41 words
A system as recited in claim 29, wherein said region of interest processing device comprises: a background-trend corrector for correcting background-trends of said regions of interest; and a pixel value normalizer for normalizing pixels of said background-trend corrected regions of interest.
Claim 35
Dependent Claim
18 words
A system as recited in claim 29, further comprising: an image thresholder for image thresholding said output images.
Claim 36
Dependent Claim
45 words
A system as recited in claim 29, wherein said cluster detector comprises: means for counting a number of clustered microcalcifications in said output image; means for determining an output image to be positive if said number of clustered microcalcifications is at least a predetermined value.
Claim 37
Dependent Claim
13 words
A system as recited in claim 32, wherein said predetermined number is two.
Claim 38
Dependent Claim
60 words
A system as recited in claim 29, wherein said neural network comprises: an input layer having a number of input units determined by a number of pixels in one half of power spectra of said regions of interest; at least one hidden layer, wherein a number of hidden layers is determined using receiver operating characteristic analysis; and an output layer.
Claim 39
Dependent Claim
89 words
A system as recited in claim 29, further comprising: said neural network having an input layer having a plurality of input units, at least one hidden layer having a plurality of groups, each of said groups having a plurality of units, an output layer, and connection weights between said layers; and connection weights between said layers being shift-invariant as w(i,j;x,y)=w(i-x, j-y) where w(i,j;x,y) denotes a connection weight between a unit at a location (i,j) in a first layer and a unit at a location (x,y) in a subsequent layer.
Claim 40
Dependent Claim
167 words
A system as recited in claim 29, further comprising: means for training said neural network using a modified error back propagation method: ##EQU9## and in the case of said output layer, δ.sub.l.sup.L (x,y)=ƒ'(N.sub.l.sup.L (x,y)+b.sub.l.sup.L) [T(x,y)-O.sub.l.sup.L (x,y)] where: N p l (x,y) is an input of the unit at said location (x,y); W p ,q l (x,y) is the connection weights; O p l (x,y) is an output of the unit at said location (x,y); l=(1,2, . . . L) is a layer number; p=(1,2, . . . P l ) is a group number in a lth layer; q=(1,2, . . . P l +1) is a group number in a (l+1)th layer; ƒ(x) is a thresholding function; ƒ'(x) is a derivative of ƒ(x); Δw p ,q 1 (x,y) is a change of the connection weight; Δb p l is a change in a bias; η is a learning rate; α is a the constant factor of a moment term; and T(x,y) is a desired output image.
Claim 41
Independent Claim
143 words
A system for detecting clustered microcalcifications in a digital mammogram, comprising: a device for obtaining a digital mammogram; a detector connected to said device which detects suspected microcalcifications in said digital mammogram; a region of interest selector, connected to said device and said detector, which selects regions of interest in said digital mammogram corresponding to said suspected microcalcifications; a region of interest processing device outputting a processed regions of interest; an input data scaler connected to said region of interest processing device for numerically scaling said processed regions of interest; a shift-invariant neural network trained to detect clustered microcalcifications receiving as input data said numerically scaled regions of interest from said data scaler and outputting corresponding output images; a microcalcification detector for detecting microcalcifications in said output images; and an apparatus for performing image feature analysis on microcalcifications detected by said microcalcification detector.
Claim 42
Independent Claim
143 words
A system for detecting clustered microcalcifications in a digital mammogram, comprising: a device for obtaining a digital mammogram; a detector connected to said device which detects suspected microcalcifications in said digital mammogram; a region of interest selector, connected to said device and said detector, which selects regions of interest in said digital mammogram corresponding to said suspected microcalcifications; a region of interest processing device outputting a processed regions of interest; an input data scaler connected to said region of interest processing device for numerically scaling said processed regions of interest; a shift-invariant neural network trained to detect clustered microcalcifications receiving as input data said numerically scaled regions of interest from said data scaler and outputting corresponding output images; a microcalcification detector for detecting microcalcifications in said output images; and an apparatus for performing image feature analysis on microcalcifications detected by said microcalcification detector.
Claim 43
Dependent Claim
41 words
A system as recited in claim 37, wherein said region of interest processing device comprises: a background-trend corrector for correcting background-trends of said regions of interest; and a pixel value normalizer for normalizing pixels of said background-trend corrected regions of interest.
Claim 44
Dependent Claim
18 words
A system as recited in claim 37, further comprising: an image thresholder for image thresholding said output images.
Claim 45
Dependent Claim
45 words
A system as recited in claim 37, wherein said cluster detector comprises: means for counting a number of clustered microcalcifications in said output image; means for determining an output image to be positive if said number of clustered microcalcifications is at least a predetermined value.
Claim 46
Dependent Claim
13 words
A system as recited in claim 40, wherein said predetermined number is two.
Claim 47
Dependent Claim
60 words
A system as recited in claim 37, wherein said neural network comprises: an input layer having a number of input units determined by a number of pixels in one half of power spectra of said regions of interest; at least one hidden layer, wherein a number of hidden layers is determined using receiver operating characteristic analysis; and an output layer.
Claim 48
Dependent Claim
88 words
A system as recited in claim 37, further comprising: said neural network having an input layer having a plurality of input units, at least one hidden layer having a plurality of groups, each of said groups having a plurality of units, an output layer, and connection weights between said layers; and connection weights between said layers being shift-invariant as w(i,j;x,y)=w(i-x,j-y) where w(i,j;x,y) denotes a connection weight between a unit at a location (i,j) in a first layer and a unit at a location (x,y) in a subsequent layer.
Claim 49
Dependent Claim
167 words
A system as recited in claim 37, further comprising: means for training said neural network using a modified error back propagation method: ##EQU10## and in the case of said output layer, δ.sub.l.sup.L (x,y)=ƒ'(N.sub.l.sup.L (x,y)+b.sub.l.sup.L) [T(x,y)-O.sub.l.sup.L (x,y)] where: N p l (x,y) is an input of the unit at said location (x,y); W p ,q l (x,y) is the connection weights; O p l (x,y) is an output of the unit at said location (x,y); l=(1,2, . . . L) is a layer number; p=(1,2, . . . P l ) is a group number in a lth layer; q=(1,2, . . . P l +1) is a group number in a (l+1)th layer; ƒ(x) is a thresholding function; ƒ'(x) is a derivative of ƒ(x); Δw p ,q l (x,y) is a change of the connection weight; Δb p l is a change in a bias; η is a learning rate; α is a the constant factor of a moment term; and T(x,y) is a desired output image.
Claim 50
Dependent Claim
31 words
A system as recited in claim 37, wherein said apparatus for performing image feature analysis comprises: a image feature extracter; and an image feature classifier connected to said image feature extracter.
Claim 51
Dependent Claim
18 words
A system as recited in claim 45, wherein said image feature classifier comprises means for performing feature thresholding.
Claim 52
Dependent Claim
63 words
A system as recited in claim 37, further comprising: a second region of interest selector connected to said microcalcification detector for selecting second regions of interest corresponding to said microcalcifications detected by said microcalcification detector; a image feature extracter for extracting image features of said microcalcifications in said second regions of interest; and an image feature classifier connected to said image feature extracter.
Claim 53
Dependent Claim
35 words
A system as recited in claim 47, further comprising: a display for displaying said digital mammogram; and wherein said image feature classifier further comprises means for indicating said detected microcalcifications on said displayed digital mammogram.
References Cited (Backward Citations)
US4907156A
Cited
US4839807A
Cited
US4841555A
Cited
US4851984A
Cited
US4875165A
Cited
US4918534A
Cited
US5072384A
Cited
US5133020A
Cited
US5150292A
Cited
US5224177A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1999-02-12 | 1999-08-14 | 2000-02-13 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2003-02-12 | 2003-08-14 | 2004-02-13 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2007-02-12 | 2007-08-14 | 2008-02-13 | $7700.00 | Window Closed / Lapsed |