Multi-modal MRI synthesization based on StarGAN
Sun Fu-hai, Xiaoying Tang · 2020
In magnetic resonance image (MRI) analysis, it is often necessary and beneficial to analyze multi-modal MRIs. However, it is typically costly to acquire images of multiple modalities. In this context, cross-modality MRI synthesization has a great potential, for which task the generative adversarial network (GAN) technique has been identified to be useful. The main limitation of GAN is that it can only transfer between two modalities and will not work if more than one modalities need to be generated from another single one. In this work, we propose to use StarGAN for multi-modal MRI synthesization. In other words, StarGAN is used to generate MRIs of multiple modalities from a single modality at one shot. In our experiment, we show that StarGAN is more time-saving than multiple GANs and also more accurate.