間質性肺病変のコンピュータ分析におけるグレイレベル・ランレングス解析の応用

概要

論文の詳細を見る
We have been developing computerized schemes using texture analysis for automated detection and characterization of interstitial infiltrates on chest radiographs. To improve the performance of the schemes, new alternative texture measures need to be developed. In this study, we applied gray-level run length analysis to digital chest x-ray images and examined its potential usefulness for the detection of interstitial infiltrates. In addition, we proposed new run length matrices, which do not depend on direction, namely, the line-component run length matrix and the area-component run length matrix. Physical texture measures were defined based on the new gray-level run length analyses, such as the total number of different line components at each gray level in the line-component run length matrix and the total number of different area components at each gray level in the area-component run length matrix. We found that gray-level nonuniformity, which shows nonuniformity of gray-level distribution in the run length matrix, was an effective measure to identify correctly most difficult normal and most difficult abnormal cases, i.e., 50% correct detection for ten difficult false-negative cases, and 18% correct detection for eleven difficult false-positive cases in the previous texture analysis. Since gray-level nonuniformity represents different properties of the interstitial infiltrate patterns compared with texture analysis based on Fourier analysis, we believe that gray-level nonuniformity provides new and alternative physical measures for the detection and characterization of interstitial infiltrates on chest radiographs. In addition, our preliminary results indicated that new texture measures which were calculated from line-or area-component run length matrices may be useful for accurate detection and characterization of interstitial infiltrates on chest radiographs.
公益社団法人日本放射線技術学会の論文
2000-03-20

間質性肺病変のコンピュータ分析におけるグレイレベル・ランレングス解析の応用

スポンサーリンク

概要

著者

関連論文

スポンサーリンク