کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

CT Super-Resolution Using Arbitrary Scale Diffusion Model

نویسندگان :

Mahsa Nadafi Ghahnavieh¹ Saeed Masoudnia² Hamid Soltanian-Zadeh³

1- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر 2- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر 3- دانشگاه تهران، دانشکده مهندسی برق و کامپیوتر

کلمات کلیدی :

Super-Resolution،Computed Tomography،Diffusion Models،Low-Resolution،Image Reconstruction

چکیده :

Computed Tomography (CT) is a critical imaging modality widely used in medical diagnostics, offering detailed visualization of anatomical structures. However, low-resolution (LR) CT images, often necessitated by reduced radiation doses, suffer from loss of information, blurred edges, and increased noise. This study introduces a novel super-resolution (SR) model based on an enhanced Implicit Diffusion Model (IDM) architecture. It utilizes a dynamic scale factor adjustment mechanism and a multi-scale LR guidance network to address arbitrary-scale SR tasks. The model achieves high-fidelity CT image reconstruction with increased flexibility and scalability by incorporating neural implicit representations and modifications to the U-Net architecture. Experimental evaluations demonstrate the superiority of our proposed network over state-of-the-art methods such as SRGAN and multi-scale attention networks, achieving Peak Signal-to-Noise Ratio (PSNR) values of 40.38, 33.07, and 29.81 at ×2, ×4, and ×8 magnifications, respectively. Structural Similarity Index Measure (SSIM) scores confirm the model's preservation of critical details and textures. The proposed method offers significant improvements in CT image quality, paving the way for its integration into clinical workflows to enhance diagnostic accuracy while maintaining patient safety.