US 5,289,374 AGranted

Method and system for analysis of false positives produced by an automated scheme for the detection of lung nodules in digital chest radiographs

AssigneeArch Development Corporation

InventorsKunio Doi, Tsuneo Matsumoto, Maryellen L. Giger, Akiko Kano

Life of this patent

Estimating the term…

Loading patent…

Abstract

A computerized method and system for reducing the number of false-positive detections of nodule candidates in the detection of abnormalities in digital chest radiography. The image is initially subjected to an image difference technique where the detection sensitivity is increased so as to avoid missing small nodules which might otherwise go undetected. Such a technique tends to increase the number of false-positives, however, leading to possible incorrect diagnoses of the radiographs. To reduce the number of false-positives, feature extraction techniques are applied to grown regions around the nodule candidates, in order to provide computer generated information concerning the candidates. A data base of parameters common to false-positives is compared to calculated parameters of a candidate of interest. The candidates with grown region parameters within the data base range common to false-positives are eliminated as being probable false-positive detections due to normal background anatomical features.

Claims12

Dependent claims

A method for reducing the number of false-positive detections in a computerized scheme for detecting true-positive abnormalities on chest radiographs, comprising the steps of:

(a) storing a plurality of predetermined threshold values in memory, wherein said plurality of predetermined threshold values are characteristics of known false-positives related to average edge-gradient, edge-gradient distribution, standard deviation of edge-gradient orientation, transition point information, size, circularity, shape irregularity and contrast values;
(b) generating a digital image of an object;
(c) subjecting said digital image to a signal-to-noise ratio (SNR) enhancement filtering process and an SNR suppressed filtering process in order to produce respective first and second filtered digital images;
(d) obtaining a difference image by subtracting said second filtered digital image from said first filtered digital image, thereby enhancing the conspicuity of abnormality candidates included on said difference image;
(e) detecting the locations of said abnormality candidates included on said difference image;
(f) calculating a first set of parameters related to size, circularity, shape irregularity, edge-gradient, and contrast values for each abnormality candidate detected on said difference image;
(g) eliminating abnormality candidates having said first set of parameters related to size, circularity, shape irregularity, edge-gradient and contrast values within predetermined value ranges corresponding to a first set of said plurality of predetermined threshold values stored in memory;
(h) performing background trend correction on said digital difference image for the abnormality candidates remaining after elimination of said abnormality candidates in step (g);
(i) calculating, for each remaining abnormality candidate, a second set of parameters related to size, circularity, shape irregularity, edge-gradient and transition point;
and
(j) eliminating, as false-positives, said remaining abnormality candidates which have at least one of said parameter values calculated in step (i) within a predetermined value range corresponding to a second set of said plurality of predetermined threshold values stored in memory, thus providing an indication of the locations of said true-positive abnormalities on said chest radiographs.

8 more elements

A system for reducing the number of false-positive detections in a computerized scheme for detecting true-positive abnormalities on chest radiographs, comprising:

means for storing a plurality of predetermined threshold values, wherein said threshold values are characteristics of known false-positives related to average edge-gradient, edge-gradient distribution, standard deviation of edge-gradient orientation, transition point information, size, circularity, shape irregularity and contrast values;
means for generating a digital image of an object;
means for subjecting said digital image to a signal-to-noise ratio (SNR) enhancement filtering process and a signal-to-noise (SNR) suppressed filtering process in order to produce respective first and second filtered digital images;
means for obtaining a difference image by subtracting said second filtered digital image from said first filtered digital image, thereby enhancing the conspicuity of abnormality candidates included on said difference image;
means for calculating a first set of parameters related to size, circularity, shape irregularity, edge-gradient and contrast values for each abnormality candidate;
means for eliminating abnormality candidates having said first set of parameters related to size, circularity, shape irregularity, edge-gradient and contrast values within predetermined value ranges corresponding to a first set of said plurality of predetermined threshold values stored in memory;
means for performing background trend correction on said digital difference image for the abnormality candidates remaining after elimination of said abnormality candidates having size, circularity, shape irregularity, edge-gradient and contrast parameters within said predetermined value ranges;
means for calculating, for each remaining abnormality candidate, a second set of parameters related to size, circularity, shape irregularity, edge-gradient and transition point;
and
means for eliminating, as false-positives, said remaining abnormality candidates which have at least one of said second set of parameters within a predetermined value range corresponding to a second set of said plurality of predetermined threshold values stored in memory, thus providing an indication of the locations of said true-positive abnormalities on said chest radiographs.

7 more elements

Dependencies are inferred from the claim text.