This thesis has created a new Snake model that overcomes many of the limitations of the traditional finite difference snake. This new deformable model combines a novel user initialization process with a finite element B-spline snake to create a powerful semi-automatic segmentation method. Using the simple but powerful initialization process, the user recognizes critical points and regions in a specified order, and transfers this knowledge to the model. By drawing lines across the object of interest, importatn information pertaining to the global shape of the object, such as width and symmetry, is imparted to the model.The snake is parameterized using minimum number of model degrees of freedom necessary and these degrees of freedom are placed in optimal positions around the object, based on the critical points and features recognized by the user via the input lines. Thus, the model is more like a deformable template than a local snake model - it is less sensitive to noise and more amenable to propagation to subsequent image slices in a volume image or time series. Unlike a traditional deformable template model however, it is constructed and positioned by the user rather than preconstructed and automatically initialized by the segmentation system. The template snake isinitialized very close to the object boundary and is very similar in shape. Furthermore, it is "aware" of its position with respect to the object. This thesis also describes the computation of the external image forces and how the known initial position and shape of the snake can be used to design object-specific image forces.