US5732697A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Shift-invariant artificial neural network for computerized detection of clustered microcalcifications in mammography
Assignee: Arch Development Corp
•
Filing: 1995-11-22
•
Granted: 1998-03-31
•
Expires: 2015-11-22
Abstract
Abstract A computerized method and system using a shift-invariant artificial neural network (SIANN) for the quantitative analysis of image data. A series of digitized medical images are used to train an artificial neural network to differentiate between diseased and normal tissue. The sum of the weights in groups between layers is constrained to be substantially zero so as to avoid saturation of layers which would otherwise be saturated by low frequency background noise. The method and system also include utilizing training-free zones to exclude from training the center portions of microcalcifications in the digitized images. The method and system further include rule-based selection criteria for providing a more accurate diagnosis.
Inventors
- 👤 Wei Zhang
- 👤 Kunio Doi
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-24 09:48
CPC Classifications
G; G06; G06T; G06T7/00; G06T7/0002; G06T7/0012
Claims Analysis
Total Claims: 99
Claim 1
Independent Claim
249 words
A method for detecting diseased tissue using a shift-invariant artificial neural network (SIANN) including an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of said at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer, the method comprising the steps of: obtaining plural digitized training medical images; extracting plural training regions of interest from the plural digitized training medical images; converting the plural training regions of interest into numerical data; assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; training the SIANN using the numerical data and plural corresponding desired outputs; obtaining a digitized patient medical image to be analyzed; and detecting diseased tissue using the trained SIANN and the digitized patient medical image.
Claim 2
Independent Claim
249 words
A method for detecting diseased tissue using a shift-invariant artificial neural network (SIANN) including an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of said at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer, the method comprising the steps of: obtaining plural digitized training medical images; extracting plural training regions of interest from the plural digitized training medical images; converting the plural training regions of interest into numerical data; assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; training the SIANN using the numerical data and plural corresponding desired outputs; obtaining a digitized patient medical image to be analyzed; and detecting diseased tissue using the trained SIANN and the digitized patient medical image.
Claim 3
Dependent Claim
72 words
The method of claim 1, wherein the step of detecting diseased tissue comprises the steps of: selecting a patient region of interest from the digitized patient medical image; converting the patient region of interest into patient numerical data; inputting the patient numerical data into the trained SIANN; outputting a filtered image at the output layer; and classifying the filtered image as representing diseased tissue or normal tissue based on the filtered image.
Claim 4
Dependent Claim
74 words
The method as claimed in claim 2, wherein the classifying step comprises: thresholding the filtered image to yield a binary image based on a threshold pixel value; tracing regions in the binary image determined to be greater than the threshold pixel value in the thresholding step; discarding noise regions from the traced regions to identify candidate regions; and classifying the filtered image as representing diseased tissue or normal tissue based on the candidate regions.
Claim 5
Dependent Claim
68 words
The method according to claim 3, wherein the step of classifying the filtered image based on the candidate regions comprises: calculating an area of each of the candidate regions; eliminating from the candidate regions any candidate region with a calculated area greater than an area threshold to identify remaining candidate regions; classifying the filtered image as representing diseased tissue or normal tissue based on the remaining candidate regions.
Claim 6
Dependent Claim
126 words
The method according to claim 4, wherein the step of classifying the filtered image as representing diseased tissue or normal tissue based on the remaining candidate regions comprises: calculating for each of the remaining candidate regions, an area of a square with an edge length corresponding to a respective longest axis of each remaining candidate region; calculating a linearity index for each of said remaining candidate regions based on the areas of the remaining candidate regions and the areas of the squares; eliminating from the remaining candidate regions any of the remaining candidate regions having a linearity index greater than a linearity threshold to identify non-linear candidate regions; and classifying the filtered image as representing diseased tissue or normal tissue based on the non-linear candidate regions.
Claim 7
Dependent Claim
36 words
The method according to claim 5, wherein the step of eliminating from the remaining candidate regions comprises: eliminating from the remaining candidate regions any of the remaining candidate regions having a linearity index greater than 0.75.
Claim 8
Dependent Claim
51 words
The method according to claim 4, wherein the step of classifying the filtered image as representing diseased tissue or normal tissue based on the remaining candidate regions comprises: classifying the filtered image as an image representing diseased tissue if a number of remaining candidate region is greater than a predetermined threshold.
Claim 9
Dependent Claim
61 words
The method according to claim 4, wherein the step of classifying the filtered image as representing diseased tissue or normal tissue based on the remaining candidate regions comprises: setting a classification threshold on two remaining candidate regions; and classifying the filtered image as an image representing diseased tissue if a number of remaining candidate regions is greater than the classification threshold.
Claim 10
Dependent Claim
27 words
The method according to claim 4, wherein the eliminating step comprises: eliminating from the candidate regions any candidate region with a calculated area greater than 100 pixels.
Claim 11
Dependent Claim
23 words
The method according to claim 3, wherein the tracing step comprises: tracing regions in the binary image using a four-connectivity region growing technique.
Claim 12
Dependent Claim
60 words
The method according to claim 3, wherein the discarding step comprises: setting a minimum area threshold; calculating an area of each of the traced regions; discarding as a noise region any of the traced regions with a calculated area less than the minimum area threshold; and identifying as a candidate region any traced region not discarded as a noise region.
Claim 13
Dependent Claim
24 words
The method according to claim 11, wherein the step of setting a minimum area threshold comprises: setting a minimum area threshold to two pixels.
Claim 14
Dependent Claim
37 words
The method according to claim 2, wherein the step of converting the patient regions of interest comprises: converting the patient regions of interest into 10-bit gray-levels; and scaling down the 10-bit gray-levels by a factor of one-hundred.
Claim 15
Dependent Claim
21 words
The method according to claim 1, wherein the training step comprises: training the SIANN using an error back propagation training method.
Claim 16
Dependent Claim
41 words
The method according to claim 1, wherein the training step comprises: using -0.9 and 0.9 as convergence targets of -1 and 1; using a learning rate, η, of 0.1; and using a constant factor of a moment term, α, of 0.9.
Claim 17
Dependent Claim
25 words
The method according to claim 1, wherein the training step comprises: training the SIANN iteratively, until a training error falls below a training error threshold.
Claim 18
Dependent Claim
71 words
The method according to claim 1, wherein the training step comprises: training the SIANN iteratively, until an area, A z , under a receiver operating characteristic (ROC) curve reaches a global maximum and begins to decrease; and reverting the values of each input of said at least one set of plural weighted inputs to their corresponding values when the area, A z , under the ROC curve was the global maximum.
Claim 19
Dependent Claim
49 words
The method according to claim 1, further comprising the step of: building the SIANN before the assigning step so that there are two hidden layers, with a first of the two hidden layers including two hidden groups and a second of the two hidden layers including one hidden group.
Claim 20
Dependent Claim
32 words
The method according to claim 1, wherein the converting step comprises: converting the plural training regions of interest into 10-bit gray-levels; and scaling down the 10-bit gray-levels by a factor of one-hundred.
Claim 21
Dependent Claim
49 words
The method according to claim 1, wherein the assigning step comprises: assigning values to each input of said at least one set of plural weighted inputs of the SIANN so that each of said units of the SIANN receives inputs from a 7×7 region of a corresponding preceding layer.
Claim 22
Independent Claim
264 words
A method for detecting diseased tissue using a shift-invariant artificial neural network (SIANN) including an input layer, an output layer, and at least one hidden, the output layer-including at least one output unit, each hidden layer of at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of the at least one output unit and each hidden unit of the at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer, the method comprising the steps of: obtaining plural digitized training medical images; extracting plural training regions of interest from the plural digitized training medical images; converting the plural training regions of interest into numerical data; assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; designating a training-free zone and plural desired outputs for each of the plural training regions of interest; training the SIANN using the numerical data and the plural corresponding desired outputs; obtaining a digitized patient medical image to be analyzed; and detecting diseased tissue using the trained SIANN and the digitized patient medical image.
Claim 23
Independent Claim
264 words
A method for detecting diseased tissue using a shift-invariant artificial neural network (SIANN) including an input layer, an output layer, and at least one hidden, the output layer-including at least one output unit, each hidden layer of at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of the at least one output unit and each hidden unit of the at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer, the method comprising the steps of: obtaining plural digitized training medical images; extracting plural training regions of interest from the plural digitized training medical images; converting the plural training regions of interest into numerical data; assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; designating a training-free zone and plural desired outputs for each of the plural training regions of interest; training the SIANN using the numerical data and the plural corresponding desired outputs; obtaining a digitized patient medical image to be analyzed; and detecting diseased tissue using the trained SIANN and the digitized patient medical image.
Claim 24
Dependent Claim
72 words
The method of claim 21, wherein the step of detecting diseased tissue comprises the steps of: selecting a patient region of interest from the digitized patient medical image; converting the patient region of interest into patient numerical data; inputting the patient numerical data into the trained SIANN; outputting a filtered image at the output layer; and classifying the filtered image as representing diseased tissue or normal tissue based on the filtered image.
Claim 25
Dependent Claim
74 words
The method as claimed in claim 22, wherein the classifying step comprises: thresholding the filtered image to yield a binary image based on a threshold pixel value; tracing regions in the binary image determined to be greater than the threshold pixel value in the thresholding step; discarding noise regions from the traced regions to identify candidate regions; and classifying the filtered image as representing diseased tissue or normal tissue based on the candidate regions.
Claim 26
Dependent Claim
68 words
The method according to claim 23, wherein the step of classifying the filtered image based on the candidate regions comprises: calculating an area of each of the candidate regions; eliminating from the candidate regions any candidate region with a calculated area greater than an area threshold to identify remaining candidate regions; classifying the filtered image as representing diseased tissue or normal tissue based on the remaining candidate regions.
Claim 27
Dependent Claim
126 words
The method according to claim 24, wherein the step of classifying the filtered image as representing diseased tissue or normal tissue, based on the remaining candidate regions comprises: calculating for each of the remaining candidate regions, an area of a square with an edge length corresponding to a respective longest axis of each remaining candidate region; calculating a linearity index for each of said remaining candidate regions based on the areas of the remaining candidate regions and the areas of the squares; eliminating from the remaining candidate regions any of the remaining candidate regions having a linearity index greater than a linearity threshold to identify non-linear candidate regions; and classifying the filtered image as representing diseased tissue or normal tissue based on the non-linear candidate regions.
Claim 28
Dependent Claim
36 words
The method according to claim 25, wherein the step of eliminating from the remaining candidate regions comprises: eliminating from the remaining candidate regions any of the remaining candidate regions having a linearity index greater than 0.75.
Claim 29
Dependent Claim
51 words
The method according to claim 24, wherein the step of classifying the filtered image as representing diseased tissue or normal tissue based on the remaining candidate regions comprises: classifying the filtered image as an image representing diseased tissue if a number of remaining candidate region is greater than a predetermined threshold.
Claim 30
Dependent Claim
61 words
The method according to claim 24, wherein the step of classifying the filtered image as representing diseased tissue or normal tissue based on the remaining candidate regions comprises: setting a classification threshold to two remaining candidate regions; and classifying the filtered image as an image representing diseased tissue if a number of remaining candidate regions is greater than the classification threshold.
Claim 31
Dependent Claim
27 words
The method according to claim 24, wherein the eliminating step comprises: eliminating from the candidate regions any candidate region with a calculated area greater than 100 pixels.
Claim 32
Dependent Claim
24 words
The method according to claim 23, wherein the tracing step comprises: tracing regions in the binary output image using a four-connectivity region growing technique.
Claim 33
Dependent Claim
60 words
The method according to claim 23, wherein the discarding step comprises: setting a minimum area threshold; calculating an area of each of the traced regions; discarding as a noise region any of the traced regions with a calculated area less than the minimum area threshold; and identifying as a candidate region any traced region not discarded as a noise region.
Claim 34
Dependent Claim
24 words
The method according to claim 31, wherein the step of setting a minimum area threshold comprises: setting a minimum area threshold to two pixels.
Claim 35
Dependent Claim
49 words
The method according to claim 21, wherein the assigning step comprises: assigning values to each input of said at least one set of plural weighted inputs of the SIANN so that each of said units of the SIANN receives inputs from a 7×7 region of a corresponding preceding layer.
Claim 36
Dependent Claim
21 words
The method according to claim 21, wherein the training step comprises: training the SIANN using an error back propagation training method.
Claim 37
Dependent Claim
41 words
The method according to claim 21, wherein the training step comprises: using -0.9 and 0.9 as convergence targets of -1 and 1; using a learning rate, η, of 0.1; and using a constant factor of a moment term, α, of 0.9.
Claim 38
Dependent Claim
25 words
The method according to claim 21, wherein the training step comprises: training the SIANN iteratively, until the training error falls below a training error threshold.
Claim 39
Dependent Claim
71 words
The method according to claim 21, wherein the training step comprises: training the SIANN iteratively, until an area, A z , under a receiver operating characteristic (ROC) curve reaches a global maximum and begins to decrease; and reverting the values of each input of said at least one set of plural weighted inputs to their corresponding values when the area, A z , under the ROC curve was the global maximum.
Claim 40
Dependent Claim
49 words
The method according to claim 21, further comprising the step of: building the SIANN before the assigning step so that there are two hidden layers, with a first of the two hidden layers including two hidden groups and a second of the two hidden layers including one hidden group.
Claim 41
Dependent Claim
32 words
The method according to claim 21, wherein the converting step comprises: converting the plural training regions of interest into 10-bit gray-levels; and scaling down the 10-bit gray-levels by a factor of one-hundred.
Claim 42
Dependent Claim
37 words
The method according to claim 21, wherein the step of converting the patient regions of interest comprises: converting the patient regions of interest into 10-bit gray-levels; and scaling down the 10-bit gray-levels by a factor of one-hundred.
Claim 43
Dependent Claim
28 words
The method according to claim 21, wherein the designating step comprises: designating a circular training-free zone with a nine pixel diameter centered about a center of a microcalcification.
Claim 44
Independent Claim
231 words
A method for training a shift-invariant artificial neural network (SIANN) to detect diseased tissue, wherein the SIANN includes an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of said at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer, the method comprising the steps of: obtaining plural digitized training medical images; extracting plural training regions of interest from the plural digitized training medical images; converting the plural training regions of interest into numerical data; assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; and training the SIANN using the numerical data and plural corresponding desired outputs.
Claim 45
Independent Claim
231 words
A method for training a shift-invariant artificial neural network (SIANN) to detect diseased tissue, wherein the SIANN includes an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of said at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer, the method comprising the steps of: obtaining plural digitized training medical images; extracting plural training regions of interest from the plural digitized training medical images; converting the plural training regions of interest into numerical data; assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; and training the SIANN using the numerical data and plural corresponding desired outputs.
Claim 46
Independent Claim
236 words
A method for training a shift-invariant artificial neural network (SIANN) to detect diseased tissue, wherein the SIANN includes an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of the at least one output unit and each hidden unit of the at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; obtaining plural digitized training medical images; extracting plural training regions of interest from the plural digitized training medical images; converting the plural training regions of interest into numerical data; assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; designating a training-free zone and plural desired outputs for each of the plural training regions of interest; and training the SIANN using the numerical data and the plural corresponding desired outputs.
Claim 47
Independent Claim
236 words
A method for training a shift-invariant artificial neural network (SIANN) to detect diseased tissue, wherein the SIANN includes an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of the at least one output unit and each hidden unit of the at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; obtaining plural digitized training medical images; extracting plural training regions of interest from the plural digitized training medical images; converting the plural training regions of interest into numerical data; assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; designating a training-free zone and plural desired outputs for each of the plural training regions of interest; and training the SIANN using the numerical data and the plural corresponding desired outputs.
Claim 48
Independent Claim
257 words
A system for detecting diseased tissue, comprising: a shift-invariant artificial neural network (SIANN) including an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group, and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of said at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; means for obtaining plural digitized training medical images; means for extracting plural training regions of interest from the plural digitized training medical images; means for converting the plural training regions of interest into numerical data; means for assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; means for training the SIANN using the numerical data and plural corresponding desired outputs; means for obtaining a digitized patient medical image to be analyzed; and means for detecting diseased tissue using the trained SIANN and the digitized patient medical image.
Claim 49
Independent Claim
257 words
A system for detecting diseased tissue, comprising: a shift-invariant artificial neural network (SIANN) including an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group, and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of said at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; means for obtaining plural digitized training medical images; means for extracting plural training regions of interest from the plural digitized training medical images; means for converting the plural training regions of interest into numerical data; means for assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; means for training the SIANN using the numerical data and plural corresponding desired outputs; means for obtaining a digitized patient medical image to be analyzed; and means for detecting diseased tissue using the trained SIANN and the digitized patient medical image.
Claim 50
Dependent Claim
79 words
The system of claim 44, wherein the means for detecting diseased tissue comprises: means for selecting a patient region of interest from the digitized patient medical image; means for converting the patient region of interest into patient numerical data; means for inputting the patient numerical data into the trained SIANN; means for outputting a filtered image at the output layer; and means for classifying the filtered image as representing diseased tissue or normal tissue based on the filtered image.
Claim 51
Dependent Claim
85 words
The system as claimed in claim 45, wherein the means for classifying comprises: means for thresholding the filtered image to yield a binary image based on a threshold pixel value; means for tracing regions in the binary image determined to be greater than the threshold pixel value used by the means for thresholding; means for discarding noise regions from the traced regions to identify candidate regions; and means for classifying the filtered image as representing diseased tissue or normal tissue based on the candidate regions.
Claim 52
Dependent Claim
75 words
The system according to claim 46, wherein the means for classifying the filtered image based on the candidate regions comprises: means for calculating an area of each of the candidate regions; means for eliminating from the candidate regions any candidate region with a calculated area greater than an area threshold to identify remaining candidate regions; and means for classifying the filtered image as representing diseased tissue or normal tissue based on the remaining candidate regions.
Claim 53
Dependent Claim
128 words
The system according to claim 47, wherein the means for classifying the filtered image based on the remaining candidate regions comprises: means for calculating for each of the remaining candidate regions an area of a square with an edge length corresponding to a respective longest axis of each remaining candidate region; means for calculating a linearity index for each of the remaining candidate regions based on the areas of the remaining candidate regions and the areas of the corresponding squares; means for eliminating from the remaining candidate regions any of the remaining candidate regions having a linearity index greater than a linearity threshold to identify non-linear candidate regions; and means for classifying the filtered image as representing diseased tissue or normal tissue based on the non-linear candidate regions.
Claim 54
Dependent Claim
38 words
The system according to claim 48, wherein the means for eliminating from the remaining candidate regions comprises: means for eliminating from the remaining candidate regions any of the remaining candidate regions having a linearity index greater than 0.75.
Claim 55
Dependent Claim
46 words
The system according to claim 47, wherein the means for classifying the filtered image based on the remaining candidate regions comprises: means for classifying the filtered image as an image representing diseased tissue if a number of remaining candidate regions is greater than a predetermined threshold.
Claim 56
Dependent Claim
58 words
The system according to claim 47, wherein the means for classifying the filtered image based on the remaining candidate regions comprises: means for setting a classification threshold to two remaining candidate regions; and means for classifying the filtered image as an image representing diseased tissue if a number of remaining candidate regions is greater than the classification threshold.
Claim 57
Dependent Claim
30 words
The system according to claim 47, wherein the means for eliminating comprises: means for eliminating from the candidate regions any candidate region with a calculated area greater than 100 pixels.
Claim 58
Dependent Claim
27 words
The system according to claim 46, wherein the means for tracing comprises: means for tracing regions in the binary output image using a four-connectivity region growing technique.
Claim 59
Dependent Claim
69 words
The system according to claim 46, wherein the means for discarding comprises: means for setting a minimum area threshold; means for calculating an area of each of the traced regions; means for discarding as a noise region any of the traced regions with a calculated area less than the minimum area threshold; and means for identifying as a candidate region any traced region not discarded as a noise region.
Claim 60
Dependent Claim
26 words
The system according to claim 54, wherein the means for setting a minimum area threshold comprises: means for setting a minimum area threshold to two pixels.
Claim 61
Dependent Claim
41 words
The system according to claim 45, wherein the means for converting the patient regions of interest comprises: means for converting the patient regions of interest into 10-bit gray-levels; and means for scaling down the 10-bit gray-levels by a factor of one-hundred.
Claim 62
Dependent Claim
24 words
The system according to claim 44, wherein the means for training comprises: means for training the SIANN using an error back propagation training method.
Claim 63
Dependent Claim
48 words
The system according to claim 44, wherein the means for training comprises: means for using -0.9 and 0.9 as convergence targets of -1 and 1; means for using a learning rate, η, of 0.1; and means for using a constant factor of a moment term, α, of 0.9.
Claim 64
Dependent Claim
28 words
The system according to claim 44, wherein the means for training comprises: means for training the SIANN iteratively, until a training error falls below a training error threshold.
Claim 65
Dependent Claim
70 words
The system according to claim 44, wherein the means for training comprises: means for training the SIANN iteratively, until an area, A z , under a receiver operating characteristic (ROC) curve reaches a global maximum and begins to decrease; and means for reverting the values of each of the plural weighted inputs to their corresponding values when the area, A z , under the ROC curve was the global maximum.
Claim 66
Dependent Claim
51 words
The system according to claim 44, further comprising: means for building the SIANN before the means for assigning assigns values so that there are two hidden layers, with a first of the two hidden layers including two hidden groups and a second of the two hidden layers including one hidden group.
Claim 67
Dependent Claim
37 words
The system according to claim 44, wherein the means for converting comprises: means for converting the plural training regions of interest into 10-bit gray-levels; and means for scaling down the 10-bit gray-levels by a factor of one-hundred.
Claim 68
Dependent Claim
52 words
The system according to claim 44, wherein the means for assigning comprises: means for assigning values to each input of said at least one set of plural weighted inputs of the SIANN so that each of said units of the SIANN receives inputs from a 7×7 region of a corresponding preceding layer.
Claim 69
Independent Claim
277 words
A system for detecting diseased tissue, comprising: a shift-invariant artificial neural network (SIANN) including an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group, and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of the at least one output unit and each hidden unit of the at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; means for obtaining plural digitized training medical images; means for extracting plural training regions of interest from the plural digitized training medical images; means for converting the plural training regions of interest into numerical data; means for assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; means for designating a training-free zone and plural desired outputs for each of the plural training regions of interest; means for training the SIANN using the numerical data and the plural corresponding desired outputs; means for obtaining a digitized patient medical image to be analyzed; and means for detecting diseased tissue using the trained SIANN and the digitized patient medical image.
Claim 70
Independent Claim
277 words
A system for detecting diseased tissue, comprising: a shift-invariant artificial neural network (SIANN) including an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group, and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of the at least one output unit and each hidden unit of the at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; means for obtaining plural digitized training medical images; means for extracting plural training regions of interest from the plural digitized training medical images; means for converting the plural training regions of interest into numerical data; means for assigning values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; means for designating a training-free zone and plural desired outputs for each of the plural training regions of interest; means for training the SIANN using the numerical data and the plural corresponding desired outputs; means for obtaining a digitized patient medical image to be analyzed; and means for detecting diseased tissue using the trained SIANN and the digitized patient medical image.
Claim 71
Dependent Claim
79 words
The system of claim 64, wherein the means for detecting diseased tissue comprises: means for selecting a patient region of interest from the digitized patient medical image; means for converting the patient region of interest into patient numerical data; means for inputting the patient numerical data into the trained SIANN; means for outputting a filtered image at the output layer; and means for classifying the filtered image as representing diseased tissue or normal tissue based on the filtered image.
Claim 72
Dependent Claim
85 words
The system as claimed in claim 65, wherein the means for classifying comprises: means for thresholding the filtered image to yield a binary image based on a threshold pixel value; means for tracing regions in the binary image determined to be greater than the threshold pixel value used by the means for thresholding; means for discarding noise regions from the traced regions to identify candidate regions; and means for classifying the filtered image as representing diseased tissue or normal tissue based on the candidate regions.
Claim 73
Dependent Claim
75 words
The system according to claim 66, wherein the means for classifying the filtered image based on the candidate regions comprises: means for calculating an area of each of the candidate regions; means for eliminating from the candidate regions any candidate region with a calculated area greater than an area threshold to identify remaining candidate regions; and means for classifying the filtered image as representing diseased tissue or normal tissue based on the remaining candidate regions.
Claim 74
Dependent Claim
128 words
The system according to claim 67, wherein the means for classifying the filtered image based on the remaining candidate regions comprises: means for calculating for each of the remaining candidate regions an area of a square with an edge length corresponding to a respective longest axis of each remaining candidate region; means for calculating a linearity index for each of the remaining candidate regions based on the areas of the remaining candidate regions and the areas of the corresponding squares; means for eliminating from the remaining candidate regions any of the remaining candidate regions having a linearity index greater than a linearity threshold to identify non-linear candidate regions; and means for classifying the filtered image as representing diseased tissue or normal tissue based on the non-linear candidate regions.
Claim 75
Dependent Claim
38 words
The system according to claim 68, wherein the means for eliminating from the remaining candidate regions comprises: means for eliminating from the remaining candidate regions any of the remaining candidate regions having a linearity index greater than 0.75.
Claim 76
Dependent Claim
46 words
The system according to claim 67, wherein the means for classifying the filtered image based on the remaining candidate regions comprises: means for classifying the filtered image as an image representing diseased tissue if a number of remaining candidate regions is greater than a predetermined threshold.
Claim 77
Dependent Claim
58 words
The system according to claim 67, wherein the means for classifying the filtered image based on the remaining candidate regions comprises: means for setting a classification threshold to two remaining candidate regions; and means for classifying the filtered image as an image representing diseased tissue if a number of remaining candidate regions is greater than the classification threshold.
Claim 78
Dependent Claim
30 words
The system according to claim 67, wherein the means for eliminating comprises: means for eliminating from the candidate regions any candidate region with a calculated area greater than 100 pixels.
Claim 79
Dependent Claim
27 words
The system according to claim 66, wherein the means for tracing comprises: means for tracing regions in the binary output image using a four-connectivity region growing technique.
Claim 80
Dependent Claim
69 words
The system according to claim 66, wherein the means for discarding comprises: means for setting a minimum area threshold; means for calculating an area of each of the traced regions; means for discarding as a noise region any of the traced regions with a calculated area less than the minimum area threshold; and means for identifying as a candidate region any traced region not discarded as a noise region.
Claim 81
Dependent Claim
26 words
The system according to claim 74, wherein the means for setting a minimum area threshold comprises: means for setting a minimum area threshold to two pixels.
Claim 82
Dependent Claim
41 words
The system according to claim 65, wherein the means for converting the patient regions of interest comprises: means for converting the patient regions of interest into 10-bit gray-levels; and means for scaling down the 10-bit gray-levels by a factor of one-hundred.
Claim 83
Dependent Claim
24 words
The system according to claim 64, wherein the means for training comprises: means for training the SIANN using an error back propagation training method.
Claim 84
Dependent Claim
48 words
The system according to claim 64, wherein the means for training comprises: means for using -0.9 and 0.9 as convergence targets of -1 and 1; means for using a learning rate, η, of 0.1; and means for using a constant factor of a moment term, α, of 0.9.
Claim 85
Dependent Claim
28 words
The system according to claim 64, wherein the means for training comprises: means for training the SIANN iteratively, until a training error falls below a training error threshold.
Claim 86
Dependent Claim
70 words
The system according to claim 64, wherein the means for training comprises: means for training the SIANN iteratively, until an area, A z , under a receiver operating characteristic (ROC) curve reaches a global maximum and begins to decrease; and means for reverting the values of each of the plural weighted inputs to their corresponding values when the area, A z , under the ROC curve was the global maximum.
Claim 87
Dependent Claim
51 words
The system according to claim 64, further comprising: means for building the SIANN before the means for assigning assigns values so that there are two hidden layers, with a first of the two hidden layers including two hidden groups and a second of the two hidden layers including one hidden group.
Claim 88
Dependent Claim
37 words
The system according to claim 64, wherein the means for converting comprises: means for converting the plural training regions of interest into 10-bit gray-levels; and means for scaling down the 10-bit gray-levels by a factor of one-hundred.
Claim 89
Dependent Claim
31 words
The system according to claim 64, wherein the means for designating comprises: means for designating a circular training-free zone with a nine pixel diameter centered about a center of a microcalcification.
Claim 90
Dependent Claim
52 words
The system according to claim 64, wherein the means for assigning comprises: means for assigning values to each input of said at least one set of plural weighted inputs of the SIANN so that each of said units of the SIANN receives inputs from a 7×7 region of a corresponding preceding layer.
Claim 91
Independent Claim
255 words
A system for training a shift-invariant artificial neural network (SIANN) to detect diseased tissue, the system comprising: a SIANN including an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of said at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; means for obtaining plural digitized training medical images; means for extracting plural training regions of interest from the plural digitized training medical images; means for converting the plural training regions of interest into numerical data; means for assigning values to each input of said at least one plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; means for designating a training-free zone and plural desired outputs for each of the plural training regions of interest; and means for training the SIANN using the numerical data and the plural corresponding desired outputs.
Claim 92
Independent Claim
255 words
A system for training a shift-invariant artificial neural network (SIANN) to detect diseased tissue, the system comprising: a SIANN including an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of said at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; means for obtaining plural digitized training medical images; means for extracting plural training regions of interest from the plural digitized training medical images; means for converting the plural training regions of interest into numerical data; means for assigning values to each input of said at least one plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; means for designating a training-free zone and plural desired outputs for each of the plural training regions of interest; and means for training the SIANN using the numerical data and the plural corresponding desired outputs.
Claim 93
Independent Claim
235 words
A system for training a shift-invariant artificial neural network (SIANN) to detect diseased tissue, the system comprising: a SIANN including an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of the at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; means for obtaining plural digitized training medical images; means for extracting plural training regions of interest from the plural digitized training medical images; means for converting the plural training regions of interest into numerical data; means for assigning values to each input of said at least one plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; and means for training the SIANN using the numerical data and plural corresponding desired outputs.
Claim 94
Independent Claim
235 words
A system for training a shift-invariant artificial neural network (SIANN) to detect diseased tissue, the system comprising: a SIANN including an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and the output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of the at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; means for obtaining plural digitized training medical images; means for extracting plural training regions of interest from the plural digitized training medical images; means for converting the plural training regions of interest into numerical data; means for assigning values to each input of said at least one plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; and means for training the SIANN using the numerical data and plural corresponding desired outputs.
Claim 95
Dependent Claim
295 words
A computer program product, comprising: a computer storage medium and a computer program code mechanism embedded in the computer storage medium for causing a computer to train a shift-invariant artificial neural network (SIANN) to detect diseased tissue, the computer program code mechanism comprising: a first computer code device configured to generate a SIANN includes an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of said at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; a second computer code device configured to obtain plural digitized training medical images; a third computer code device configured to extract plural training regions of interest from the plural digitized training medical images; a fourth computer code device configured to convert the plural training regions of interest into numerical data; a fifth computer code device configured to assign values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; and a sixth computer code device configured to train the SIANN using the numerical data and plural corresponding desired outputs.
Claim 96
Dependent Claim
295 words
A computer program product, comprising: a computer storage medium and a computer program code mechanism embedded in the computer storage medium for causing a computer to train a shift-invariant artificial neural network (SIANN) to detect diseased tissue, the computer program code mechanism comprising: a first computer code device configured to generate a SIANN includes an input layer, an output layer, and at least one hidden layer, the output layer including at least one output unit, each hidden layer of the at least one hidden layer including at least one group and each group including at least one hidden unit, wherein the input layer is connected to a first hidden layer of said at least one hidden layer, and output layer is connected to a last hidden layer of said at least one hidden layer, and wherein each output unit of said at least one output unit and each hidden unit of said at least one hidden unit in the SIANN receives at least one set of plural weighted inputs from a corresponding preceding layer; a second computer code device configured to obtain plural digitized training medical images; a third computer code device configured to extract plural training regions of interest from the plural digitized training medical images; a fourth computer code device configured to convert the plural training regions of interest into numerical data; a fifth computer code device configured to assign values to each input of said at least one set of plural weighted inputs so that for each set of said at least one set of plural weighted inputs a sum of all of said values in said each set is substantially zero; and a sixth computer code device configured to train the SIANN using the numerical data and plural corresponding desired outputs.
Claim 97
Dependent Claim
48 words
The computer program product as claimed in claim 87, further comprising: a seventh computer code device configured to obtain a digitized patient medical image to be analyzed; and an eighth computer code device configured to detect diseased tissue using the trained SIANN and the digitized patient medical image.
Claim 98
Dependent Claim
33 words
The computer program product as claimed in claim 87, further comprising: a seventh computer code device configured to designate a training-free zone in at least one of the plural training regions of interest.
Claim 99
Dependent Claim
70 words
The computer program product as claimed in claim 87, further comprising: a seventh computer code device configured to obtain a digitized patient medical image to be analyzed; an eighth computer code device configured to detect diseased tissue using the trained SIANN and the digitized patient medical image; and a ninth computer code device configured to designate a training-free zone in at least one of the plural training regions of interest.
References Cited (Backward Citations)
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2001-03-30 | 2001-09-29 | 2002-03-31 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2005-03-30 | 2005-09-29 | 2006-03-31 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2009-03-30 | 2009-09-29 | 2010-03-31 | $7700.00 | Window Closed / Lapsed |