Segmentation of chest radiographs

Segmentation of chest radiographs is a necessary step before any computer — aided diagnosis method are applied. This restricts the region to be searched for particular diseases (nodules, interstitial disease etc.). It may also help in detecting the cardiac abnormalities (cardiomegaly). Before applying the segmentation function, preprocessing routines remove the unexposed borders and standardize the position of the image. The analysis is performed in several steps. First, the histogram analysis finds a threshold value, that eliminates the densest parts of the image and permits some contour points to be detected. Based on these points, a Sobel gradient image is analysed to remove noise attached to the lungs. Then, the abdominal and mediastinum contour is searched by analysing profiles perpendicular to the bottom of the lungs and the spine, respectively. The algorithm has been tested on a pilot study of clinical images with a good rate of correctly segmented lungs.