Lung Segmentation on Postero-anterior Digital Chest Radiographs Using Active Contours Isaac Iglesias 1 , Pablo G. Tahoces 2 , Miguel Souto 3 , Anxo Mart´ ınez de Alegr´ ıa 3 , Mar´ ıa J. Lado 4 , and Juan J. Vidal 3 1 Laboratory for Radiologic Image Research, University of Santiago, 15782 Santiago isaacigl@usc.es 2 Department of Electronics and Computer Science. University of Santiago Campus Sur s/n, 15782 Santiago 3 Department of Radiology Complejo Hospitalario Universitario de Santiago (CHUS) University of Santiago, 15782 Santiago 4 Department of Computer Science, University of Vigo Campus Lagoas-Marcosende, 36200 Vigo Abstract. A computerized pulmonary segmentation based on the de- tection of oriented edges was performed in postero-anterior (PA) digital radiography (DR) images. To further improve detection of lung contours, a method based on the use of active contours models was developed. First, the technique calculates a set of reference lines to determine the relative position of the lungs in the image. Then, vertical and horizontal rectangular regions of interest (ROIs) are studied to identify the prelimi- nary edge. These points are an approximation to the lung edges that are adjusted using the active contours models. We studied the influence of the different parameters of the active contours on the final result over 30 DR images. Results prove that the active contour models, with selected parameters, can be used to improve the results of a given segmentation scheme. 1 Introduction Digital radiography (DR) is becoming the standard for pulmonary imaging [1] [2]. DR provides high spatial and contrast resolution for the lung parenchyma. DR also opened the possibilities for computer-aided diagnosis (CAD) applications in chest imaging, such as pulmonary nodules detection [2] [3], characterization of interstitial disease [4], measurement of pulmonary volume [5], and detection of cardiomegaly [6] and pneumotoraces [7]. A survey of the CAD in the chest radiography can be found in [8]. Lung segmentation is a necessary prerequisite for all of these quantitative analysis applications. Several investigators have developed techniques for CAD segmentation of pulmonary DR images. Xu and Doi [9] have determined the rib cage boundary using a set of ROIs, which are selected over the maxima and min- ima of the gray-level values on horizontal profiles. Duryea et al. [10] searched the A. Fred et al. (Eds.): SSPR&SPR 2004, LNCS 3138, pp. 538–546, 2004. c  Springer-Verlag Berlin Heidelberg 2004