The efficient extraction of the cryoablation iceball from a time series of 3D images is crucial during cryoablation to assist the interventionalist in determining the coverage of the tumor by the ablated volume. Conventional semi-automatic segmentation tools such as ITK-SNAP and 3D Slicer's Fast Marching Segmentation can attain accurate iceball segmentation in retrospective studies, however, they are not ideal for intraprocedure real time segmentation, as they require time-consuming manual operations, such as the input of fiducials and the extent of the segmented region growth. In this paper, we present an innovative approach for the segmentation of the iceball during cryoablation, that executes a fully automatic computation. Our approach is based on the graph cuts segmentation framework, and incorporates prior information of iceball shape evolving in time, modeled using experimentally-derived iceball growth parameters. Modeling yields a shape prior mask image at each timepoint of the imaging time series for use in the segmentation. Segmentation results of our method and the ITK-SNAP method are compared for 8 timepoints in 2 cases. The results indicate that our fully automatic approach is accurate, robust and highly efficient compared to manual and semi-automatic approaches.