Seed Extension Based Interactive Medical Volume Segmentation Method

This paper proposes an interactive segmentation method based on seed extension to tackle the problems of the mincut/max-flows algorithm, which was extensively validated for many interactive segmentation applications. The extension is performed by constructing minimum spanning forests (MSF) from seed voxels imposed by users, which minimizes the weights of edges from the seeds to segmented lines (cuts). Compared with the graph cuts-based method, the proposed method segments the volume image into more than two regions of interests. Moreover, the proposed method performs 10 times faster when segmenting volumes composed of more than 240 slices, as the time complexity of constructing MSF is quasi-linear, whereas the min-cut/max-flow is polynomial. Keywords-component; interactive segmentation; medical image processing; graph cuts; minimum spanning forests