Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

3D Brain Generation using Auto-Encoding Generative Adversarial Networks with Cycle Consistent Embedding

Xing, Shibo (2022) 3D Brain Generation using Auto-Encoding Generative Adversarial Networks with Cycle Consistent Embedding. Master's Thesis, University of Pittsburgh. (Unpublished)

Download (1MB) | Preview


An array of generative adversarial networks (GANs) have been accomplishing the realistic generation of full 3D brain images. This largely follows a common procedure of sampling from a latent space prior Z (i.e., random vectors) and mapping it to realistic images in X (e.g., 3D brains), but a naıve implementation also comes with the ubiquitous mode collapse issue. This challenge has recently been addressed by strongly imposing certain characteristics, such as Gaussianness, to the prior by also explicitly mapping X to Z via encoder. This Auto- Encoder type GANs, however, fail to accurately map 3D brain images to the desirable prior, which the generator assumes to be sampling the random vectors from. While Variational Auto-Encoding GAN (VAE-GAN) handles this mode collapse issue by explicitly imposing Gaussianness, this also causes blurriness in images. In this thesis, we demonstrate how our cycle consistent embedding GAN (CCE-GAN) is able to solve both the mode collapse and blurriness issues by accurately encoding 3D MRIs to the standard normal prior while maintaining the image generation quality. Using our trained novel model with T1 MRI brain images from Alzheimer’s Disease Neuroimaging Initiative (ADNI) and FLAIR tumor MRI brain images from Brain Tumor Segmentation (BraTS) datasets, we will show how an improved prior Z space can lead to an output distribution free of mode collapse and of high image quality. We also quantitatively and qualitatively assess the embeddings to reaffirm the importance of embedding in GAN for 3D brain generation.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Xing, Shiboshx26@pitt.edushx26
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairHwang, Seong
Committee MemberKovashka, Adriana
Committee MemberDavneet,
Date: 2 June 2022
Date Type: Publication
Defense Date: 14 December 2021
Approval Date: 2 June 2022
Submission Date: 3 May 2022
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Number of Pages: 19
Institution: University of Pittsburgh
Schools and Programs: School of Computing and Information > Computer Science
Degree: MS - Master of Science
Thesis Type: Master's Thesis
Refereed: Yes
Uncontrolled Keywords: Auto-Encoder, Latent Space, Generative Adversarial Network, Cycle Consistency, 3D MRI
Related URLs:
Date Deposited: 02 Jun 2022 21:10
Last Modified: 02 Jun 2022 21:10


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item