MS-PRNet: A Novel Hierarchical Framework for Optimized and Precise Medical Image Registration

Medical image registration faces challenges due to limited effective receptive fields and inadequate understanding of global spatial correspondences in traditional Convolutional Neural Networks (CNN). The existing methods, while addressing these issues, also introduced problems, such as increased model complexity, which results in higher computational costs and re-duced real-time performance. In response, we propose the Multi-Scale Pyramid Registration Network (MS-PRNet), which not only effectively handles large-scale displacements but also avoids the increased computational complexity associated with existing methods. MS-PRNet utilizes a hierarchical scaling method to process image pairs, calculating displacement deformation fields at multiple levels and integrating them to produce accurate registration outcomes. Tested on the LPBA40 and MindbogglelOl datasets, MS-PRNet outperformed existing methods, particularly excelling in regions with complex anatomical structures. This approach improves diagnostic precision and enhances clinical applications without compromising computational efficiency.