sTBI-GAN: An adversarial learning approach for data synthesis on traumatic brain segmentation

Xiangyu Zhao, Di Zang, Sheng Wang, Zhenrong Shen, Kai Xuan, Zeyu Wei, Zhe Wang, Ruizhe Zheng, Xuehai Wu, Zheren Li, Qian Wang, Zengxin Qi, Lichi Zhang

Abstract: Highlights•Introduce sTBI-GAN for augmenting traumatic brain segmentation data using image inpainting.•Develop image inpainting for simultaneous grayscale MR images and label map synthesis.•Adopt a coarse-to-fine synthesis strategy with edge information to enhance TBI image textures.•Develop a registration-based template pipeline to diversify synthesized brain trauma variance.•sTBI-GAN shows superior data augmentation and effectively aids segmentation in experiments.