SA-GAN: Structure-Aware GAN for Organ-Preserving Synthetic CT Generation

In medical image synthesis, model training could be challenging due to the inconsistencies between images of different modalities even with same patient, typically caused by internal status/tissue changes as are usually obtained at a time. This paper proposes novel deep learning method, Structure-aware Generative Adversarial Network (SA-GAN), that preserves shapes and locations in-consistent structures when generating images. SA-GAN is employed generate synthetic computed tomography (synCT) from magnetic resonance imaging (MRI) two parallel streams: global stream translates input MRI CT domain while local automatically segments inconsistent organs, maintains their in MRI, organ intensities CT. Through extensive experiments on pelvic dataset, we demonstrate provides clinically acceptable accuracy both synCTs segmentation supports MR-only treatment planning disease sites status changes.