Automated selection of regions of interest for quantitative analysis of lung textures in digital chest radiographs

Abstract

In order to implement a computerized scheme for quantitative analysis of interstitial lung disease in chest radiographs in clinical situations, a fully automated method of selecting many square regions of interest (ROIs) in peripheral lung areas are developed. First, the peripheral lung regions are identified, based on the automated detection of lung apices, ribcage edges, and diaphragm. Then a large number of ROIs are selected sequentially by filling in the identified peripheral regions. Finally, those ROIs containing sharp edges are removed based on an edge gradient analysis, for which a gradient-weighted edge orientation histogram is employed. Approximately 200-400 ROIs were automatically selected for each case with this method. The evaluation of using ROC analysis indicated that the automated ROI selection method was effective in quantitative analysis of lung textures in digital chest radiographs.