SALSA - A Simple Automatic Lung Segmentation Algorithm

Abstract

The accurate segmentation of pulmonary tissue is of great importance for several diagnostic tasks. A simple and fast algorithm for performing lung segmentation is proposed here. The method combines several simple image processing operations to achieve the nal segmentation and can be divided into two problems. The srt is the lung segmentation, that identi es regions such as backgroung, trachea, vessels, and left and right lungs, and it is complicated by the presence of noise, artifacts, low contrast and diseases. The second is the lobe segmentation, where the left lung is divided into two lobes, the upper and lower lobes, and the right into three lobes, the upper, middle and lower lobes. This second problem is harder due to the fact that the membranes dividing the lobes, the pleurae, are very thin and are not clearly visualized in the computerized tomography exams, besides the possible occurence of lobectomies (surgical lobe removal), diseases that may degrade the image qulaity, or noise during image acquisition. Both methods were developed in order to produce an authomatic method, and we have already produced results similar to the ones produced by other studies for the rst problem, using the testing methodology of the lung segmentation challenge LOLA11.