Mask-aware transformer with structure invariant loss for CT translation

Wenting Chen, Wei Zhao, Zhen Chen, Tianming Liu, Li Liu, Jun Liu, Yixuan Yuan

Abstract: Highlights•The proposed MAFormer presents the first work to exploit subtraction image for CT translation.•The proposed MATNorm combines the subtraction image with plain CT image through transformer.•To preserve the CT structure, we introduce a structure invariant loss between plain and synthetic CT images.•We manually collect MPECT dataset with complete four phases, i.e., plain, arterial, portal venous and delayed phases.•Quantitative and qualitative results prove the effectiveness of our method and its superiority to current methods.