In this paper, a translation-based cross-modality deformable medical image registration model is proposed. It focuses on preserving spatial correlation among local and global features of both modalities. This model uses a discriminator-free StyleGAN2 as the translation network and a U-Net-inspired architecture as the registration network to generate a deformation field that will warp the moving image to a fixed image using a spatial transformer network (STN). This registration network includes a CNN-based local and a BGRU-based global feature extraction module, a transformer-based local-global spatial correlation module, and a novel super-resolution loss function to register finer-level lymph node-like structures properly. The proposed model is evaluated on two pelvis datasets for MRI to CT registration. Experiments show a 36.7% increase in training speed, a 5.40% increase in structural similarity index, a 29.85% increase in normalized cross-correlation coefficient, and a 2.19% decrease in mean-squared error for cross-modality image registration compared to state-of-the-art translation-based registration models. This registration model has broader applications in multimodality image segmentation, lymph node classification, etc.