Cited time in webofscience Cited time in scopus

Conditional Generative Adversarial Network for Predicting 3D Medical Images Affected by Alzheimer’s Diseases

Conditional Generative Adversarial Network for Predicting 3D Medical Images Affected by Alzheimer’s Diseases
Jung, EuijinLuna, Acevedo Miguel AndresPark, Sang Hyun
DGIST Authors
Park, Sang Hyun
Issued Date
Predicting the evolution of Alzheimer’s disease (AD) is important for accurate diagnosis and the development of personalized treatments. However, learning a predictive model is challenging since it is difficult to obtain a large amount of data that includes changes over a long period of time. Conditional Generative Adversarial Networks (cGAN) may be an effective way to generate images that match specific conditions, but they are impractical to generate 3D images due to memory resource limitations. To address this issue, we propose a novel cGAN that is capable of synthesizing MR images at different stages of AD (i.e., normal, mild cognitive impairment, and AD). The proposed method consists of a 2D generator that synthesizes an image according to a condition with the help of 2D and 3D discriminators that evaluate how realistic the synthetic image is. We optimize both the 2D GAN loss and the 3D GAN loss to determine whether multiple consecutive 2D images generated in a mini-batch have real or fake appearance in 3D space. The proposed method can generate smooth and natural 3D images at different conditions by using a single network without large memory requirements. Experimental results show that the proposed method can generate better quality 3D MR images than 2D or 3D cGAN and can also boost the classification performance when the synthesized images are used to train a classification model. © 2020, Springer Nature Switzerland AG.
Springer Science and Business Media Deutschland GmbH
Related Researcher
  • 박상현 Park, Sang Hyun 로봇및기계전자공학과
  • Research Interests 컴퓨터비전; 인공지능; 의료영상처리
Files in This Item:

There are no files associated with this item.

Appears in Collections:
Department of Robotics and Mechatronics Engineering Medical Image & Signal Processing Lab 2. Conference Papers


  • twitter
  • facebook
  • mendeley

Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.