Abstract #0823

Water and Fat Separation with a Dixon Conditional Generative Adversarial Network (DixonCGAN)

Jong Bum Son¹, Ken-Pin Hwang¹, Marion E. Scoggins², Basak E. Dogan³, Gaiane M. Rauch², Mark D. Pagel⁴, and Jingfei Ma¹

¹Imaging Physics Department, The University of Texas MD Anderson Cancer Center, Houston, TX, United States, ²Diagnostic Radiology Department, The University of Texas MD Anderson Cancer Center, Houston, TX, United States, ³Department of Diagnostic Radiology, The University of Texas Southwestern Medical Center, Dallas, TX, United States, ⁴Cancer Systems Imaging Department, The University of Texas MD Anderson Cancer Center, Houston, TX, United States

A Dixon conditional generative adversarial network (DixonCGAN) was developed for Dixon water and fat separation. For the robust water image reconstruction, DixonCGAN performs water and fat separation with three processing steps: (1) phase-correction with DixonCGAN, (2) error-correction for DixonCGAN processing, and (3) the final water and fat separation. A conditional generative adversarial network (CGAN) originally designed to change photo styles could be successfully modified to perform phase-correction with improved global and local image details. Moreover, localized deep-learning processing errors could be effectively recovered with the proposed deep-learning error-correction processes.

This abstract and the presentation materials are available to members only; a login is required.

Join Here