The 3D medical image segmentation problem typically involves assigning labels to 3D pixels, called voxels, which comprise a given medical volume. In its simplest form the segmentation problem involves assigning the labels "part of the structure of interest" or "not part of the structure" to each voxel using locally measured properties and prior knowledge of human anatomy. Robust segmentation remains an open research problem today due to the significant challenges in the task including: partial volume averaging, overlapping intensity distributions and image noise. In the face of these challenges prior knowledge needs to be added to make the segmentation methods more robust. Active contours were introduced in the late 1980's mainly to address situations in which the object to be segmented had a single closed boundary. To address situations in which the object(s) to be segmented have unknown topology the level set framework was recently introduced to segment medical images. Unlike active contours, the level set method relies on an implicit shape representation rather than an explicit shape representation and hence new methods to impose prior knowledge about expected shape have to be devised for the new framework. This paper explores recent segmentation methods from four research groups which address the task of imposing prior knowledge of shape for object boundary segmentation. Three of the methods impose priors onto the level set technique and one employs a medial axis shape representation and Statistical shape information to guide a model-based segmentation. All of the methods include a notion of a statistical shape distribution. Each method is described, analyzed for its strengths and weaknesses. The paper concludes with a comparison of all four methods and recommendations for their applicability.