We review the normal-based interpolating subdivision scheme proposed in [1]. We show that it allows the user to exactly represent circles/spheres whenever suitable initial data are provided, and we also prove that it enjoys the property of similarity invariance. In summary, we show that it satisfies all the requirements for the construction of a deformable model to be used in the delineation of biomedical images. We then also present experimental examples dealing with the delineation of 2D and 3D biological structures.