3

I'm considering using GANs for medical image denoising, based on previous literature, like this and this. My input to the GAN would be a high-noise image and my ideal output would be a low-noise, high-quality image.

Is the GAN architecture better suited for applications where the inputs are just random noise? Is the discriminator necessary in this case or is it better to just use a Deep CNN/Autoencoder? How do I justify using a GAN for my application?

nbro
  • 39,006
  • 12
  • 98
  • 176
Jan
  • 31
  • 3
  • 2
    Could you please share your "previous literature", so that we can get an idea of what you are trying to do? I guess something like a CycleGAN might be useful for you, where you can translate one "style" of image into another. (see [here](https://machinelearningmastery.com/what-is-cyclegan/) for example) Maybe in your case the styles could be "low noise images" and "high-noise images". How big is your dataset? You would need a large amount of images with and without noise, but they don't necessarily need to be paired. – Mafu Jan 29 '21 at 13:20
  • @Mafu I was looking at [this](https://ieeexplore.ieee.org/document/8340157) and [this](https://ieeexplore.ieee.org/document/7934380) as previous literature. I have a dataset with around 10000 paired patches (128x128) and I could acquire more if required. Thank you for your comment and please do let me know if I can provide any other information, I am currently reading more about CycleGANs. – Jan Jan 29 '21 at 13:34

0 Answers0