SYMMETRIC REGISTRATION OF MULTI-MODAL MEDICAL IMAGES BASED ON ANATOMICAL INFORMATION

This paper introduces a multi-modal symmetric image registration method leveraging the advanced SymReg-GAN architecture. The framework is specifically designed to address the challenges of aligning medical images from different modalities, such as CT and MRI, which often vary significantly in appearance while representing the same anatomical structures. The proposed method incorporates an Anatomical Attention Module (AAM), which focuses on preserving the anatomical coherence of key structures during the registration process. This ensures that critical regions are accurately aligned without losing important contextual information. Furthermore, the architecture integrates multi-scale feature extraction, enabling the model to capture both fine-grained and large-scale anatomical details across different resolution levels. By utilizing pre-trained models, the system benefits from enhanced feature representation and improved generalization, allowing for more robust performance on various datasets. A composite loss function is employed to optimize the registration process, balancing multiple objectives, including intensity similarity, structural alignment, and transformation regularization. This ensures not only precise alignment but also the generation of realistic and anatomically plausible transformations. The method's efficacy is demonstrated through extensive experiments on multi-modal abdominal datasets, where it consistently outperforms state-of-the-art techniques in terms of accuracy and robustness. The results highlight its potential to significantly advance medical image analysis, facilitating better diagnosis and treatment planning. This work contributes to the field by offering a comprehensive solution for multi-modal image registration, addressing both the complexity of medical imaging and the need for reliable, anatomically coherent alignment methods.