Lung segmentation relied on the grayscale contrast between lung parenchyma and surrounding tissues. After a segmented position of the trachea was established, two seed points were located in the right and left lungs. Using a two- and three-dimensional region-growing algorithm, the points were expanded to segment both lungs. A histogram-based threshold technique was used to remove blood vessels, and a rolling ball algorithm was applied to the lung boundary. The lung was then subdivided into 8 × 8 texture blocks. A multidimensional feature vector composed of 25 different texture features was computed and used to differentiate and group each texture block. These features included histogram statistics (ie, mean, skewness, deviation, variance, kurtosis), co-occurrence matrices (ie, energy, inertia difference, correlation, average difference, entropy difference, inertia, entropy, average sum, entropy sum), and run-length matrices (ie, long run emphasis, run length nonuniformity, low gray-level run emphasis, short run low gray-level emphasis, long run low gray-level emphasis, short run high gray-level emphasis, long run high gray-level emphasis, short run emphasis, run gray-level nonuniformity, run percentage, high gray-level run emphasis). In the training stage, the computer program learned the texture patterns of normal lung and regions affected by ILD through regions manually identified by an expert radiologist. A support vector machine was generated to distinguish normal and fibrotic lung.21 In the classification stage, the computer program automatically grouped the segmented lung as normal or fibrotic lung using the support vector machine. Grading measured disease severity based on percentage of affected regions, calculated as volume of diseased region divided by volume of segmented lung.