Diffusion Model for Generating Synthetic Contrast Enhanced CT from Non-Enhanced Heart Axial CT Images

Victor Rogério Sousa Ferreira, Anselmo Cardoso de Paiva, Aristófanes Corrêa Silva, João Dallyson Sousa de Almeida, Geraldo Braz Júnior, Francesco Renna

Abstract: This work proposes the use of a deep learning-based adversarial diffusion model to address the translation of contrast-enhanced from non-contrast-enhanced computed tomography (CT) images of the heart. The study overcomes challenges in medical image translation by combining concepts from generative adversarial networks (GANs) and diffusion models. Results were evaluated using the Peak signal to noise ratio (PSNR) and structural index similarity (SSIM) to demonstrate the model’s effectiveness in generating contrast images while preserving quality and visual similarity. Despite successes, Root Mean Square Error (RMSE) analysis indicates persistent challenges, highlighting the need for continuous improvements. The intersection of GANs and diffusion models promises future advancements, significantly contributing to clinical practice. The table compares CyTran, CycleGAN, and Pix2Pix networks with the proposed model, indicating directions for improvement.