Selective segmentation provides the opportunity to monitor and assess closely the activities of cells within a confined section of a given cell image. Segmentation of cells through traditional graph cut method is well known; however, when a selective segmentation is desired, the task becomes cumbersome. In order to take advantage of the optimum segmentation of graph cut while simultaneously adopting a local (selective) segmentation strategy, a third term is added to the graph cut energy function. The term is referred to as adaptive distance penalties (ADP). ADP constrains graph cut to any selected region of interest while using a reference image to achieve a semi-automatic segmentation. Experiments show that for fairly homogeneous cell images, the proposed model outperforms the grab cut selective segmentation.