Purpose/Significance To systematically review the current state, frontier progress and core challenges of generative models in medical image analysis, and to provide references for relevant research. Method/Process By using the literature review method, the fundamental principles, technical evolution, advantages and disadvantages of mainstream generative models represented by generative adversarial network(GAN), variational autoencoder(VAE) and diffusion models are systematically elaborated. Through the key application tasks such as cross-modal image synthesis, data augmentation, reconstruction and denoising, super-resolution, segmentation and detection, current research is summarized and classified. The model performance evaluation framework is sorted out, and a multidimensional evaluation system from technical indicators to clinical application efficacy is summarized. Result/Conclusion Generative models demonstrate great potential and application value in the field of medical image analysis, but their clinical transformation still faces challenges such as insufficient controllability and interpretability of the model, need for improved generalization and robustness, data ethics issues, and high computational overhead,etc.