Method and system for the segmentation of lung regions in lateral chest radiographs

Abstract

A method and system for the automated segmentation of the lung regions in lateral chest radiographs. This is achieved according to the invention by providing an improved computerized, automated method for image segmentation based on gray-level threshold analysis. A unique method for identifying an approximate outer bounds on the extent of the lung fields in the image is performed to restrict the region further analyzed. An iterative global gray-level thresholding method is applied based on the features of a global gray-level histogram. Features of the regions in a binary image constructed at each iteration are identified and subjected to a modified analysis to exclude regions external to the lung field. The initial lung region contour that results from this global process is used to facilitate a novel adaptive local gray level thresholding method. Individual regions-of-interest (ROIs) are placed along the initial contour. The dimensions of the several ROIs are based upon the patient anatomy enclosed therein. A unique procedure is implemented to determine the single gray-level threshold to be applied to the pixels within the individual ROIs. A composite binary image results, and a final contour is constructed to enclose “on” regions thereof. Smoothing processes are applied, including a unique adaptation of a rolling ball method, and fitted polynomial curves are spliced into the final contour.