eduzhai > Applied Sciences > Engineering >

GANDALF Generative Adversarial Networks with Discriminator-Adaptive Loss Fine-tuning for Alzheimers Disease Diagnosis from MRI

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Positron Emission Tomography (PET) is now regarded as the gold standard forthe diagnosis of Alzheimer s Disease (AD). However, PET imaging can beprohibitive in terms of cost and planning, and is also among the imagingtechniques with the highest dosage of radiation. Magnetic Resonance Imaging(MRI), in contrast, is more widely available and provides more flexibility whensetting the desired image resolution. Unfortunately, the diagnosis of AD usingMRI is difficult due to the very subtle physiological differences betweenhealthy and AD subjects visible on MRI. As a result, many attempts have beenmade to synthesize PET images from MR images using generative adversarialnetworks (GANs) in the interest of enabling the diagnosis of AD from MR.Existing work on PET synthesis from MRI has largely focused on ConditionalGANs, where MR images are used to generate PET images and subsequently used forAD diagnosis. There is no end-to-end training goal. This paper proposes analternative approach to the aforementioned, where AD diagnosis is incorporatedin the GAN training objective to achieve the best AD classificationperformance. Different GAN lossesare fine-tuned based on the discriminatorperformance, and the overall training is stabilized. The proposed networkarchitecture and training regime show state-of-the-art performance for three-and four- class AD classification tasks.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×