Paper Info
Title
Alaysis of image features of histograms of edge gradient for false positive reduction in lung nodule detection in chest radiographs
Abstract
A computer-aided diagnosis (CAD) scheme could improve radiologists' diagnostic performance in their detection of lung nodules on chest radiographs if the computer output were used as a "second opinion." The current CAD scheme that we have developed achieved a performance of 70% sensitivity and 1.7 false positives per image for our database. This database consisted of two hundred PA chest radiographs, including 100 normals and 100 abnormals (containing 122 confirmed nodules). Our purpose in this study was to improve our scheme further by incorporating new features derived from analysis of the histogram of radial edge gradients on nodule candidates. For this study, we examined a total of 426 candidates, which included 86 (70%) of 122 detected nodules and 340 non-nodules (1.7 false positives per image, for a total of 200 images) resulted from the current scheme. Approximately 80% of these false positives were due to rib-rib or rib-vessel crossings, and also to interactions between ribs and soft tissues, such as breast, cardiac, or diaphragm shadows. A 64 X 64-pixel region of interest (ROI) centered at the candidate location was selected first. The contrast of the ROI was improved by a two-dimensional background subtraction. A nodule shape matched filter was used for enhancement of the nodular pattern located in the central area of the ROI. We obtained a histogram of accumulated edge gradients as a function of the radial angles. Analysis of the histogram resulted in seven features, including the maximum, minimum, width, and standard deviation of the histogram in a selected range of radial angles. The histogram from a "true" nodule ROI tends to have a narrow, prominent peak with a large maximum value near the radial axis. However, the rib structures generally broaden the corresponding histogram, thus resulting in a large width and a high minimum value. Features derived from the histogram can be used for identifying some subtle and difficult false positives that can not be eliminated by our current CAD scheme. The rule-based test, by combining all seven features, eliminated 138 (40%) of 340 false positives without any loss of nodules. Application of an artificial neural network (ANN) removed an additional 8% of the remaining false positives with a reduction of 5% of true nodules. Our preliminary results indicate that, with this new technique, the performance of our CAD scheme could be improved further.
Year
DOI
Venue
1998
10.1117/12.310908
PROCEEDINGS OF THE SOCIETY OF PHOTO-OPTICAL INSTRUMENTATION ENGINEERS (SPIE)
Keywords
DocType
Volume
lung nodule,false positive,computer-aided diagnosis (CAD),histogram analysis,edge gradients
Conference
3338
ISSN
Citations 
PageRank 
0277-786X
11
1.07
References 
Authors
0
5
Name
Order
Citations
PageRank
xinwei xu1111.07
Shigehiko Katsuragawa217226.20
K Ashizawa3358.11
Heber MacMahon420231.61
K Doi513526.74