In modern lung cancer radiotherapy, it is important to have a precise knowledge of the real-time lung tumor position during the treatment delivery. For a real-time 3D lung imaging algorithm from a single X-ray projection image, the inaccuracies contributed by the image registration process affects much on the PCA modelling and construction process. We utilize 3 deformable image registration algorithms: Optical Flow method, Demons method and Levelset method to evaluate the effect. By making quantitative analysis and qualitative analysis, we get the conclusion: Optical Flow method works much better in registration and PCA modelling.