AIGAN: Attention-encoding Integrated Generative Adversarial Network for the reconstruction of low-dose CT and low-dose PET images

Yu Fu, Shunjie Dong, Meng Niu, Le Xue, Hanning Guo, Yanyan Huang, Yuanfan Xu, Tianbai Yu, Kuangyu Shi, Qianqian Yang, Yiyu Shi, Hong Zhang, Mei Tian, Cheng Zhuo

Abstract: Highlights•A generative adversarial network AIGAN is proposed for low-dose CT and PET image reconstruction.•The cascade generator allows the low-dose images to be enhanced twice with attention-encoding.•The dual-scale discriminator plays the zero-sum game with the cascade generator for two rounds.•The multi-scale spatial fusion module exploits inter- and intra-slice structural information.•Clinical reading and cross-domain generalization validation demonstrate the potential of AIGAN.