Accurate automatic localisation of lung nodules using Graph Cut and snakes algorithms

Lung nodule segmentation is the first and the most difficult step in every Computer Aided Diagnosis (CAD). Difficulty arises due to the boring and time-consuming nature of the manual lung segmentation process. In this paper, we propose a novel automatic lung segmentation method for accurate localization of the lung nodules in computer tomography (CT) images. We present a combination of the Graph Cut and active contour (Snakes) modeling application for CT scan image segmentation. The first step in the automatic algorithm is the enhancement of contrast and removal of noise by the Median Filter. Subsequently, lungs are segmented by active contours as ROI and next, a graph-cut method initialized by a threshold, is used to obtain more robust results. Finally, an automatic segmentation strategy is presented. We evaluated the segmentation accuracy of our method on several real and simulated nodules. In fact, 27 CT images inside the image set of the Lung Image Database Consortium (LIDC), supplied by National Center Institute (NCI), are used in our evaluations. Experimental results showed high accuracy rate and low time consumption in automatically locating the lung nodules in comparison with two existing methods and radiologists' diagnosis.