hongweilibran / diamondgan

the released codes for DiamondGAN - a scalable multiple-to-one domain translation approach in medical imaging

DiamondGAN: Unified Multi-Modal Adversarial Networks for MRI Sequences Synthesis

Q. 1: what are the applications that DiamondGAN can be used for?

Answer: With the projects with our colaborators in TUM hosipital, we used DiamondGAN for synthesizing missing modalities for MS and glioma patients. As mentioned in the miccai paper, the double inversion recovery (DIR) can be synthesized/well-reconstructed using FLAIR, T1 and T2.

Q. 2: Does DiamondGAN promise to genereate expected pathological information such as the texture of lesion or tumor?

Answer: you can combine as much information as possible from multiple sources. The hypothesis behind DiamondGAN is that the information of a target modality is distributed among other input modalities. Please see our experiment about the reconstruction of T1 constrast which requires a paramagnetic contrast agent. We somewhat failed to synthesize T1 contrast modality.

diamondgan's Issues

Hi，What is the tf version?

rt

Is starGAN?

It is very similar starGAN (figure, loss....). Have you compare your method and stargan?

Dataset directory

Hi, I could not get dataset directory. What does traiS1, trainS2 specify? Thanks