International Peer-Reviewed JournalOpen AccessISSN 2456-8880
irejournals@gmail.com+91-7433024337

Home / Current Issue / Paper 1722523

1722523 Vol 7 · Issue 3 Download Paper

Evaluating Perturbation Resilience of GAN and Diffusion Architectures for Synthetic Medical Imaging

Johan Bergstrom Aditi Rao Pedro Alvarez

Subject area: Science,Engineering and Technology  ·  Area of research: Synthetic Medical Imaging

Abstract

This paper evaluates how gracefully progressive GANs and latent diffusion models tolerate input perturbations when synthesizing medical images. Using CheXpert radiographs and OASIS brain MRI, we subject both families to Gaussian, salt-and-pepper, and FGSM perturbations and measure fidelity, perceptual distance, and generative recall. Latent diffusion consistently demonstrates higher resilience, retaining usable structure at perturbation levels where the GAN's samples fragment.

How to cite this paper

Johan Bergstrom, Aditi Rao, Pedro Alvarez "Evaluating Perturbation Resilience of GAN and Diffusion Architectures for Synthetic Medical Imaging" Iconic Research And Engineering Journals Volume 7 Issue 3 2023 Page 901-905
Johan Bergstrom, Aditi Rao, Pedro Alvarez "Evaluating Perturbation Resilience of GAN and Diffusion Architectures for Synthetic Medical Imaging" Iconic Research And Engineering Journals, vol. 7, no. 3, Sep. 2023
Johan Bergstrom, Aditi Rao, Pedro Alvarez (2023). Evaluating Perturbation Resilience of GAN and Diffusion Architectures for Synthetic Medical Imaging. Iconic Research And Engineering Journals, 7(3).
Johan Bergstrom, Aditi Rao, Pedro Alvarez "Evaluating Perturbation Resilience of GAN and Diffusion Architectures for Synthetic Medical Imaging" Iconic Research And Engineering Journals, vol. 7, no. 3, Sep. 2023.
@article{1722523,
      author = {Johan Bergstrom, Aditi Rao, Pedro Alvarez},
      title = {Evaluating Perturbation Resilience of GAN and Diffusion Architectures for Synthetic Medical Imaging},
      journal = {Iconic Research And Engineering Journals},
      year = {2023},
      volume = {7},
      number = {3},
      pages = {901-905},
      issn = {2456-8880},
      url = {https://www.irejournals.com/formatedpaper/1722523.pdf},
      abstract = {This paper evaluates how gracefully progressive GANs and latent diffusion models tolerate input perturbations when synthesizing medical images. Using CheXpert radiographs and OASIS brain MRI, we subject both families to Gaussian, salt-and-pepper, and FGSM perturbations and measure fidelity, perceptual distance, and generative recall. Latent diffusion consistently demonstrates higher resilience, retaining usable structure at perturbation levels where the GAN's samples fragment.},
      month = {September},
  }