An Automatic Method for Lung Segmentation in Thin Slice Computed Tomography Based on Random Walks

In this paper, a prior knowledge guided random walks is proposed. We combine the entropy rate superpixels with dyadic discrete wavelet transform to automatic obtain the coarse area of seeds and non-seeds by rapidly accounting the location of the lung parenchyma and the background in the anatomy. After random walks performed, a curvature-based approach is followed for amending the segmented lung contour. Experiments on a validation database consisting of 23 chest CT scans suggested that the proposed method was superior to other similar methods for lung segmentation on CT scans.