Merlin: A Vision Language Foundation Model for 3D Computed Tomography