<p>Medical image registration plays an essential role in the clinical applications of disease diagnosis, treatment, and prognosis. However, registration is often modeled as an ill-posed problem, and its feasibility is greatly reduced in the presence of large deformations, especially when no prior information is available. In order to address the aforementioned challenges, an anatomical prior guided coarse-to-fine 3D medical image registration model for large deformation is proposed. Specifically, a recursive cascaded multi-resolution registration network (RCMR) employs the coarse-to-fine strategy to progressively refine the deformation field, making large deformation registration less ill-posed. Meanwhile, the anatomical prior guided module (APGM) is proposed to serve as an anatomical prior regularization for the registration model, facilitating the feasibility of large deformation registration. Experiments are conducted on the public datasets NLST, and the experimental results demonstrate that our method achieves more accurate, robust, and topology-preserving registration performance for 3D large deformation image compared to the state-of-the-art methods.</p>