Medical Image Synthesis Based on Spatial and Frequency Domain Attention Framework

Medical image synthesis technology, which converts one type of medical image to another, plays a crucial role in improving diagnostic accuracy, optimizing treatment plans, and advancing medical research. However, existing deep learningbased models often struggle to preserve high-frequency texture information, affecting the quality of generated images and their clinical applicability. To address this, we propose a novel method that combines wavelet and spatial dual attention mechanisms. The spatial attention mechanism focuses on local details, accurately capturing subtle lesions and structural changes, while the wavelet attention mechanism preserves high-frequency textures through wavelet decomposition, enhancing the realism and detail representation of the images. Additionally, we designed a spatialfrequency domain feature fusion module that integrates information from both domains, ensuring the generated images excel in both global consistency and local details. Experimental results demonstrate that our model outperforms state-of-the-art models on multiple MRI-CT datasets, achieving significant improvements in various evaluation metrics, thus validating the superiority of our approach.