ScholarWorks@동국대학교

초록

Artificial intelligence is playing an increasingly important role in medical image generation, contributing to improved image quality and data diversity. These advances are helping to solve key problems in the clinical setting, such as low-quality scans and data sparsity. Initially, convolutional neural networks (CNNs) and generative adversarial networks (GANs) were mainly utilized, but in recent years, transformer-based models and diffusion models have emerged, leading to technological advances. These latest techniques are being actively researched for their potential for real-world clinical applications. In this paper, AI-based medical image generation techniques are comprehensively investigated, focusing on three major modalities: chest X-ray (CXR), computed tomography (CT), and magnetic resonance imaging (MRI). To organize the research, it is divided into two pillars: image enhancement and image synthesis. Image enhancement focuses on directly improving the diagnostic quality of images, including noise removal, blur correction, resolution enhancement, and restoration of incomplete data. Image synthesis, on the other hand, involves techniques that generate new data or improve clinical visibility by augmenting data, restoring anatomical structures, removing occlusions, etc. This paper analyzes the evolution of the model architecture of each approach, as well as the accessibility of the datasets used, code disclosure, quantitative metrics, and clinical evaluation cases involving radiologists. It also diagnoses the current state of medical image generation technologies and discusses future research directions, focusing on key challenges that need to be addressed in the future, such as domain generalization, establishing qualitative evaluation schemes, and integration into clinical workflows.

키워드