%0 Journal Article %T 生成模型在胸部X射线图像中的应用综述
A Review of the Application of Generative Models in Chest X-Ray Images %A 储欢 %A 肖琪 %A 王慧玲 %J Computer Science and Application %P 287-300 %@ 2161-881X %D 2025 %I Hans Publishing %R 10.12677/csa.2025.154101 %X 生成模型在医学影像领域的快速发展为胸部X射线(CXR)图像的合成、编辑与增强等领域提供了新的技术手段。本文系统综述了生成对抗网络(GAN)、变分自编码器(VAE)及扩散模型在CXR图像中的研究进展与应用。VAE通过隐变量学习生成数据分布，在疾病检测与图像重构中表现稳健，但生成图像常存在模糊现象；GAN凭借其高真实感图像生成能力，被广泛用于解决数据稀缺问题及跨模态图像合成，但其训练不稳定性和模式崩溃问题仍需优化；扩散模型凭借逐步去噪的生成机制，在图像质量与多样性上展现出超越GAN的潜力，成为当前研究热点。文章进一步分析了该领域的研究现状，总结了生成模型在数据增强、图像生成、及图像编辑中的创新应用，并对比了不同技术的优势与局限性。尽管生成模型在提升诊断效率与数据隐私保护方面成果显著，但仍面临伦理、法律及模型泛化性等挑战。未来研究需聚焦多模态生成、隐私保护框架设计及病理特征解耦，以推动生成模型在临床中的实际应用。本文为医学影像领域的研究者提供了技术参考与方向指引，具有重要的学术价值与应用前景。
The rapid development of generative models in the field of medical imaging has provided new technical means for the synthesis, editing and enhancement of chest X-ray (CXR) images. This paper systematically reviews the research progress and applications of generative adversarial networks (GANs), variational autoencoders (VAEs) and diffusion models in CXR images. VAEs generate data distributions through latent variable learning, and are robust in disease detection and image reconstruction, but the generated images are often blurred; GANs are widely used to solve data scarcity problems and cross-modal image synthesis due to their ability to generate highly realistic images, but their training instability and mode collapse problems still need to be optimized; diffusion models have the potential to surpass GANs in image quality and diversity due to their step-by-step denoising generation mechanism, and have become a current research hotspot. This paper further analyzes the current research status in this field, summarizes the innovative applications of generative models in data enhancement, image generation, and image editing, and compares the advantages and limitations of different technologies. Although generative models have achieved remarkable results in improving diagnostic efficiency and protecting data privacy, they still face challenges such as ethics, law, and model generalization. Future research needs to focus on multimodal generation, privacy protection framework design, and pathological feature decoupling to promote the practical application of generative models in clinical practice. This article provides technical references and direction guidance for researchers in the field of medical imaging, and has important academic value and application prospects. %K VAE， %K GAN， %K 扩散模型， %K 胸部X射线
VAE %K GAN %K Diffusion Model %K Chest X-Ray %U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=112590