基于双域变化编解码网络的CT截断伪影去除算法

CT Truncation Artifact Removal Algorithm Based on Dual-Domain Change Encoding and Decoding Network

CT的内重建问题对于医学影像是一个长期的挑战,由于拍摄时视场较小,会让CT图像上出现截断伪影,它们会严重降低图像质量并影响诊断。因此,本文提出了一种基于双域变换的编解码网络(Dual-Domain U-Net)模型,这个模型能够同时对投影域和图像域进行参数的调整,增强了模型的鲁棒性和保真性。本文在美国医学物理学家协会(AAPM)和国家癌症研究所的癌症成像档案馆(TCIA)的20个公共数据集上验证了模型的有效性。实验结果表明,本文提出的模型可以去除CT图像上的截断伪影,并且在图像细节保留方面的表现也更加优秀。

The interior reconstruction problem in CT has been a long-standing challenge in medical imaging.Due to the small field of view during acquisition,truncation artifacts may appear on CT images,significantly reducing image quality and impacting diagnosis.Therefore,we propose a dual-domain transformation encoding-decoding network model(Dual-domain U-Net).This model can simultaneously adjust parameters in both the projection domain and the image domain,enhancing the robustness and fidelity of the model.We validated the effectiveness of the model on 20 public datasets from the Cancer Imaging Archive(TCIA)of the American Association of Physicists in Medicine(AAPM)and the National Cancer Institute.Experimental results demonstrate that our proposed model can remove truncation artifacts on CT images and exhibit superior performance in preserving image details.