Files

Abstract

A method and system for automated classification of distinction between normal lungs and abnormal lungs with interstitial disease, based on the analysis of predetermined physical texture measures and also on a data base for normal lungs of these texture measures. The texture measures selected are the RMS variation, R, and the first moment of power spectrum, M, for lung texture. These two texture measures are normalized by using the data base for normal lungs. A single texture index is determined from the two normalized texture measures by taking into account the distribution (or the data base) of texture measures obtained from abnormal lungs, in order to facilitate the automated classification of normal and abnormal lungs. A threshold texture index is then chosen for initial selection of "abnormal" regions of interest (ROIs), which contain a large texture index above the threshold level. The selected abnormal ROIs are then subjected to three independent tests for a (1) definitely abnormal singular pattern, (2) localized abnormal pattern for two or more abnormal clustered ROIs, and (3) diffuse abnormal pattern for more than four abnormal ROIs distributed through the lung. A chest image containing any one of these abnormal patterns is classified as showing an abnormal lung with interstitial disease.

Details

Actions

PDF

from
to
Export
Download Full History