RED-Net: Radiomics-Enhanced Diffusion Network for MRI-to-PET Cross-Modality Image Synthesis

Despite PET's diagnostic potential, its widespread use is hindered by high costs and the risks associated with radiation exposure, limiting its accessibility in clinical practice. To address these challenges, synthesizing PET-like images from MRI data has emerged as a viable alternative. However, existing synthesis approaches have largely overlooked the inclusion of additional quantitative information, for example, radiomics which capture intricate tissue characteristics through quantitative metrics. In this study, we propose the Radiomics-Enhanced Diffusion Network (RED-Net) to synthesize PET images. By integrating radiomics features into the diffusion process, our network effectively captures both structural and functional information while enhancing the diagnostic relevance of the generated PET images. Experimental results demonstrate the robustness of RED-Net and its ability to achieve reliably synthesized PET images with high quality.