Liver Region Multimodal Medical Image Registration Based on Multi-Resolution VoxelMorph Network

Liver image registration plays a crucial role in the integration and comparison of different imaging modalities, particularly in clinical diagnosis, treatment planning, and ablation surgery. Due to the complex anatomical structure and varying pathological changes in the liver region, achieving high-precision registration remains a significant challenge. To address this issue, we propose a medical image registration method based on a multi-resolution VoxelMorph network. This method leverages a multi-level image pyramid strategy and incorporates Dice loss, mutual information (MI) loss, and gradient regularization during training, which enhances registration performance, especially in the liver region. This approach has important implications for clinical diagnosis and treatment planning. We used a dataset provided by Harbin Medical University for experimental evaluation, focusing on CT and MRI images. By adopting customized loss functions and multi-resolution strategies, the proposed approach achieves quantitative improvements of 2.67% in DSC and 1.88% in SSIM compared to the standard VoxelMorph method. Although there is room for improvement compared to existing methods, our approach demonstrates meaningful progress in liver region registration. It is expected to provide precise and efficient support for intraoperative real-time guidance.