Dual-Branch Score-Based Diffusion Model Incorporating Radiomics Into Synthesizing PET From MR Images

Motivation: PET imaging offers valuable diagnostic insights but is limited by high costs and radiation exposure. Synthesizing PET-like images from MRI offers a viable alternative, though current methods largely overlook quantitative metrics such as radiomics, which could enhance diagnostic accuracy. Goal(s): Our goal was to enhance the fidelity and diagnostic relevance of synthesized PET images by incorporating radiomics features into the MRI-to-PET synthesis process. Approach: We introduced a dual-branch diffusion model for MRI-to-PET synthesis, capturing both structural and functional information, with radiomics features integrated. Results: The model excelled in generating high-quality PET images, showing enhanced accuracy and preservation of diagnostic details over previous methods. Impact: The proposed method integrates radiomics into the synthesis process, enhancing the clinical utility of MRI-based PET alternatives by producing reliable, radiation-free PET-like images with improved diagnostic fidelity and accessibility.