Application of neural networks as an aid in medical diagnosis and general anomaly detection

• US 5,331,550 A
• Filed: 02/11/1993
• Issued: 07/19/1994
• Est. Priority Date: 03/05/1991
• Status: Expired due to Fees

• Current Assignee: Direct Radiography Corp. (Hologic Incorporated)
• Original Assignee: E. I. du Pont de Nemours and Company (Corteva, Inc.)
• Inventors: Mickewich, Daniel J., Beutel, Jacob, Stafford, Richard G.
• Primary Examiner(s): HUNTLEY, DAVID
• Application Number:
  US08/016,343
• Time in Patent Office: 523 Days
• Field of Search: 364/413.13, 364/413.01, 364/413.02, 364/413.14, 382/6, 382/14, 382/15, 382/27, 382/30, 382/54, 395/21, 395/23, 395/924, 395/22
• US Class Current: 382/128
• CPC Class Codes: G06F 18/2433 Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/048 Activation functions
• CPC Class Codes: G06T 2207/30068 Mammography; Breast
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06V 10/25 Determination of region of interest [ROI] or a volume of interest [VOI]
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 2201/03 Recognition of patterns in medical or anatomical images
• CPC Class Codes: G16H 15/00 ICT specially adapted for medical reports, e.g. generation or transmission thereof
• CPC Class Codes: G16H 30/40 for processing medical images, e.g. editing
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: Y10S 128/925 Neural network
• CPC Class Codes: Y10S 706/924 Medical

Abstract

A method for computer-aided detection of anomalies in an image comprise the steps of: (1) dividing the image into a plurality of m×n regions; (2) subtracting the background from each of the regions; (3) for each of the regions, selecting a smaller p×q subregion; (4) normalizing the p×q subregion; (5) feeding the p×q subregions into a neural network system, the neural network system having plural member neural networks, each trained to recognize a particular preselected anomaly type; (6) comparing each output value of the plurality of member neural networks to a first threshold; (7) selecting a maximum value from the output values which are greater than the first threshold; (8) comparing the maximum value to a second threshold above which the presence of an anomaly is indicated, and storing the result; (9) clustering a plurality of the stored results to form clusters; and (10) marking the location of the clusters.

Claims

1. A method for computer-aided detection of possible anomalies in a digitized image comprising a plurality of M×

N picture elements each representing an optical density in the digitized image, the method comprising the steps of;

a) subdividing the digitized image into a plurality of predetermined regions each comprising m×

n picture elements, where m<

M and n<

N;

b) subtracting background from each predetermined region of the digitized image;

c) selecting a subregion comprising p×

q picture elements, where p<

m and q<

n;

d) normalizing the image data from the p×

q subregion;

e) using a neural network system, analyzing each predetermined subregion of the digitized image to recognize any pattern indicative of an occurrence of a possible anomaly, the neural network system comprising at least two member neural networks each trained to recognize a particular predetermined anomaly type within a predetermined size range and to produce an output signal value indicative of the presence of said predetermined anomaly type;

f) comparing each of the output values of each of the member neural networks to a first predetermined threshold value corresponding to each member neural network above which the presence of a possible anomaly is indicated;

g) comparing each output value that exceeds each first predetermined threshold to each of the other output values that exceed each corresponding first predetermined threshold to select the maximum signal value;

h) comparing the maximum signal value to a second predetermined threshold value above which the presence of a possible anomaly is indicated;

i) based upon the comparisons of step h), determining the location within the digitized image of each possible anomaly;

j) using a clustering analysis on the locations of possible anomalies of step i), identifying each vicinity on the digitized image at which a cluster of locations of possible anomalies occurs; and

,

k) creating a marker for each cluster, having a contour that surrounds all of the anomalies in each cluster, corresponding to the digitized location of each cluster.

2. The method of claim 1 wherein the step k) itself comprises the step of placing the marker on a transparency overlay sheet.

3. The method of claim 1 wherein each subregion of p×

q picture elements is selected from the center of each m×

n region.

4. The method of claim 1 wherein the step b) comprises using a high pass spatial frequency filter.

5. The method of claim 1 wherein the step d) comprises using a two-stage normalization, the first stage being a scale invariance transformation, and the second stage being a dynamic range normalization.

6. The method of claim 1 wherein the step j) comprises using a two-stage clustering algorithm, the first stage being a microclustering binding step, and the second stage being a macroclustering binding step.

7. The method of claim 1 wherein each first threshold value of step f) is different for each member neural network.

8. The method of claim 1 wherein each first threshold value of step f) is the same for each member neural network.