Cross-modality image translation: CT image synthesis of MR brain images using multi generative network with perceptual supervision

Xianfan Gu, Yu Zhang, Wen Zeng, Sihua Zhong, Haining Wang, Dong Liang, Zhenlin Li, Zhanli Hu

Abstract: Highlights•We propose a structural aligned MRI/CT domain transformation in a cross-modal generative network.•The proposed method adopts the hybrid method of cycle consistency and style transfer in cross-domain transformation.•The proposed method leverages the joint supervision of discriminative perceptual and frozen pre-trained perceptual to enforce the structural consistency of the MRI/CT domain.•The proposed U-Net architecture combines the residual connection and U-Net connection, which reserves both the fusion of high-level features and spatial information of input images during image encoding to prevent reconstruction degradation in the upsample stage. The following experimental results validate the reconstruction quality of the generator network.