A multimodal comparison of latent denoising diffusion probabilistic models and generative adversarial networks for medical image synthesis

Table 1 Numerical evaluation of Medfusion’s autoencoder reconstruction quality. This overall confirmed the numerically measured high reconstruction quality but revealed some dataset-specific reconstruction errors (Fig. This demonstrated that Medfusion’s image quality could be further enhanced by making use of a better autoencoding architecture. Table 2 Numerical evaluation of Stable Diffusion’s autoencoder reconstruction quality. In the CRXDX dataset, Medfusion achieved a mean AUROC value of 0.57 compared to cGAN’s 0.50 (P = 0.01).