Premium
Two‐dimensional extrapolation methods for texture analysis on CT scans
Author(s) -
Sensakovic William F.,
Starkey Adam,
Armato Samuel G.
Publication year - 2007
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.2760307
Subject(s) - extrapolation , texture (cosmology) , artificial intelligence , mathematics , texture filtering , deconvolution , pattern recognition (psychology) , image texture , fractal , computer vision , computer science , image processing , statistics , image (mathematics) , mathematical analysis
The application of texture analysis to medical images may require the calculation of texture descriptors on regions of interest (ROIs) that are not completely filled by the tissue under analysis. If a texture descriptor is calculated using such “deficient” ROIs, the accuracy and computational speed may be adversely affected. This study applied 198 texture descriptors from five texture classes (first‐order statistical, second‐order statistical, Fourier, fractal, and Laws’ filtered) to lung parenchyma ROIs automatically extracted from the thoracic CT scans of ten patients. Statistically significant differences in the values of 138 of these texture descriptors were demonstrated when calculated on deficient ROIs. Three extrapolation methods (mean fill, tiled fill, and CLEAN deconvolution) then were applied to correct the deficient ROIs. Texture descriptor values were calculated and compared for the original, deficient, and corrected ROIs (based on the three extrapolation methods). Each extrapolation method induced statistically significant improvements in texture descriptor accuracy for some subset of texture descriptors. CLEAN deconvolution improved the greatest number of descriptors, demonstrated the best overall accuracy, and created ROIs that were visually most similar to the original ROIs.