摘要生成式网络在医学影像领域展现出良好应用前景，可有效缓解医学图像数据稀缺与标注困难问题。梳理了近年来主流生成式网络在多模态影像合成、低剂量图像重建及结构保持等方面的代表性工作，包括变分自动编码器、生成对抗网络与去噪扩散概率模型等框架的演进及其在医学图像生成中的应用，并对当前存在的挑战及未来研究方向进行了探讨，以期为医学图像生成方法的临床转化提供技术支持。

abstractsThe generative network shows great potential for application in the field of medical imaging, as it can effectively address the scarcity of medical image data and the difficulty of labeling. In this review, the representative studies on mainstream generative networks in multimodal image synthesis, low-dose image reconstruction and structure preservation in recent years were reviewed. This includes an overview of the evolution of frameworks such as variational autoencoder, generative adversarial network and denoising diffusion probabilistic model, as well as their application in medical image generation. The current challenges and future research directions were also discussed in order to provide technical support for the clinical implementation of medical image generation methods.