Score-based Diffusion Model을 이용한 의료 이미지 처리의 강화법

Abstract

Generative models can be very useful in the field of medical imaging. These models can be used to address data imbalance issues or to transform to different modalities. Additionally, 3D generation can be applied to clinical research, distribution analysis and more. However, medical images are more complex than natural images, making generation difficult. This means that a lot of effort is needed to create plausible generation and that it is challenging for generative models to excel in the medical image field. Nonetheless, recent advances in diffusion models have made it possible to generate high-quality images, and the use of latent diffusion models has also solved the issue of generation speed. Therefore, this paper proposes experiments on generation using diffusion models, data augmentation through generation, image-to-image transformation, 3D generation, and predicted generation. The results of this study can significantly impact the field of medical imaging by providing more accurate and comprehensive diagnostic tools for medical professionals. The use of diffusion models can also reduce the time and effort required for medical image generation and improve the overall quality of medical images, leading to better treatment outcomes for patients. This paper provides a comprehensive overview of the technical implementation and clinical applications of score-based diffusion models in medical imaging and highlights their potential to revolutionize the field of medical imaging diagnosis.