Computer aided diagnosis of interstitial lung disease: A texture feature extraction and classification approach
Published
Journal Article
An approach for the classification of normal or abnormal lung parenchyma from selected regions of interest (ROIs) of chest radiographs is presented for computer aided diagnosis of interstitial lung disease (ILD). The proposed approach uses a feed-forward neural network to classify each ROI based on a set of isotropic texture measures obtained from the joint grey level distribution of pairs of pixels separated by a specific distance. Two hundred ROIs, each 64 X 64 pixels in size (11 X 11 mm), were extracted from digitized chest radiographs for testing. Diagnosis performance was evaluated with the leave-one-out method. Classification of independent ROIs achieved a sensitivity of 90% and a specificity of 84% with an area under the receiver operating characteristic curve of 0.85. The diagnosis for each patient was correct for all cases when a 'majority vote' criterion for the classification of the corresponding ROIs was applied to issue a normal or ILD patient classification. The proposed approach is a simple, fast, and consistent method for computer aided diagnosis of ILD with a very good performance. Further research will include additional cases, including differential diagnosis among ILD manifestations. ©2003 Copyright SPIE - The International Society for Optical Engineering.
Full Text
Duke Authors
Cited Authors
- Vargas-Voracek, R; McAdams, HP; Floyd, CE
Published Date
- December 1, 1998
Published In
Volume / Issue
- 3338 /
Start / End Page
- 1502 - 1509
International Standard Serial Number (ISSN)
- 0277-786X
Digital Object Identifier (DOI)
- 10.1117/12.310882
Citation Source
- Scopus