Textural analysis of diffused disease in the lung

A method for creating a diagnostic evaluation for usual interstitial pneumonia (UIP) comprising:

• obtaining a first plurality of series of high resolution computed tomography (HRCT) slices of a plurality of subjects' lungs wherein some voxels of some of the first plurality of series indicate the presence of UIP;
• obtaining an identification of UIP voxels in at least some of the plurality of series and an identification of non-UIP voxels in at least some of the plurality of series wherein the UIP voxels represent the presence of UIP in an area of the lung represented by the UIP voxels and non-UIP voxels represent the absence of UIP in an area of the lung represented by the non-UIP voxels;
• extracting a first plurality of features from said UIP voxels and said non-UIP voxels, wherein extracting comprises measuring each of the first plurality of features as exhibited by the UIP voxels and by the non-UIP voxels and the first plurality of features comprise first-order statistics, second-order statistics, gray level co-occurrence matrix, gray level run-length matrix, and location features;
• selecting features wherein selecting comprises performing a first classifier to obtain a rank of importance for each of the plurality of features in which a higher-ranked feature provides a measure that is more accurate in differentiating UIP voxels from non-UIP voxels than is a lower-ranked feature, and eliminating one or more redundant features wherein a redundant feature comprises a feature whose measure as exhibited by voxels is highly correlated with the measure of another feature as exhibited by voxels having a higher rank of importance than the redundant feature;
• and
• constructing a predictive model wherein constructing comprises obtaining a second plurality of series of HRCT slices of subjects' lungs wherein some voxels of some of the plurality of series indicate the presence of UIP and training a number of the subset of features to provide a probability that a voxel signifies the presence of UIP by performing a second classifier on the number of the subset of features.