Interactive Segmentation of Lung Nodules using AdaBoost and Graph Cuts

In this paper, we propose an interactive method for lung nodule segmentation. Given a seed point, the segmentation process consisting of three steps is done automatically. The first step is intensity normalization. The second one is to build an energy function for graph cuts. The third one is to do the segmentation by graph cuts. In the third step, if there are imperfects in the result, we provide an interactive way to correct. The main advantages of our method are that: 1) object intensity is estimated based on energy function; 2) boundary energy of graph cuts is determined by AdaBoost; 3) a novel effective correction way is provided. Experimental results show that 1) and 2) improved the segmentation accuracy a lot, and 3) provided an effective way to correct the segmentation results even in very difficult cases.