eduzhai > Applied Sciences > Engineering >

AdaIN-Switchable CycleGAN for Efficient Unsupervised Low-Dose CT Denoising

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: Recently, deep learning approaches have been extensively studied for low-doseCT denoising thanks to its superior performance despite the fast computationaltime. In particular, cycleGAN has been demonstrated as a powerful unsupervisedlearning scheme to improve the low-dose CT image quality without requiringmatched high-dose reference data. Unfortunately, one of the main limitations ofthe cycleGAN approach is that it requires two deep neural network generators atthe training phase, although only one of them is used at the inference phase.The secondary auxiliary generator is needed to enforce the cycle-consistency,but the additional memory requirement and increases of the learnable parametersare the main huddles for cycleGAN training. To address this issue, here wepropose a novel cycleGAN architecture using a single switchable generator. Inparticular, a single generator is implemented using adaptive instancenormalization (AdaIN) layers so that the baseline generator converting alow-dose CT image to a routine-dose CT image can be switched to a generatorconverting high-dose to low-dose by simply changing the AdaIN code. Thanks tothe shared baseline network, the additional memory requirement and weightincreases are minimized, and the training can be done more stably even withsmall training data. Experimental results show that the proposed methodoutperforms the previous cycleGAN approaches while using only about half theparameters.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×