Abstract: Although various segmentation or cutout schemes have been well-developed for both 2D and 3D images, it remains a challenging problem to cutout an object from a volume dataset in an intuitive and convenient fashion.Most of existing volume cutout algorithms operate on 2D slice images, and extend the extracted information to the 3D counter.In this paper, we propose a novel volume object cutout algorithm which allows users to select 3D points by drawing strokes onto the volume rendering images.A set of 3D points is collected by casting through the drawn strokes, followed by a refinement process based on the k-mean clustering method to obtain high confidential result.Taking these refined points as seed points, the graph cut algorithm is employed to accurately and quickly extract the intended object.Experimental results demonstrate the effectiveness of our algorithm.