Numerical simulations, such as blood flow or coil deployment in an intra-cranial aneurism, are very sensitive to the boundary conditions given by the surface of the vessel walls. Despite the undisputable high quality of 3D vascular imaging modalities, artifacts and noise still hamper the extraction of this surface with enough accuracy. Previous studies took the a priori that a homogeneous object was considered to make the reconstruction from the Xray images more robust. Here, an active surface approach is described, that does not depend on any particular image similarity criterion and grounds on high speed computation of the criterion derivatives. Mean square error and normalized cross-correlation are used to successfully demonstrate our algorithm on real images acquired on an anthropomorphic phantom. Preliminary results of coil deployment simulation are also given.