A deep learning-based ring artifact correction method for X-ray CT

• Published in: Radiation detection technology and methods 2021-12, Vol.5 (4), p.493-503
• Main Authors: Yuan, Lulu, Xu, Qiong, Liu, Baodong, Wang, Zhe, Liu, Shuangquan, Wei, Cunfeng, Wei, Long
• Format: Article
• Language: English
• Subjects: Beam Physics; Hadrons; Heavy Ions; Nuclear Energy; Nuclear Physics; Particle Acceleration and Detection; Physics; Physics and Astronomy; Review
• ISSN: 2509-9930; 2509-9949
• DOI: 10.1007/s41605-021-00286-1

Purpose In X-ray CT systems, ring artifacts caused by the nonuniform response of detector elements degrades the reconstruction quality and affects the subsequent processing and quantitative analysis of the image. Method In this paper, a novel method is proposed to remove the ring artifacts in CT image by applying deep learning algorithm based on convolutional neural network (CNN) and recurrent neural network (RNN). First, the reconstructed CT images is transformed into polar coordinate system to make rings appear as stripes. Then, a CNN is constructed to detect the stripes, and a RNN is utilized to process the line artifact correction. After that, by retransforming the corrected image from polar coordinate system to Cartesian coordinate system, a ring artifact removal image can be achieved. Results The presented method can successfully reduce the CT ring artifact on simulated and real data. Specifically, in the experiment with real water phantom, the center and peripheral standard deviations reduced 46% and 24%, respectively. Conclusions The proposed method is potential to be widely deployed in industrial and medical CT systems, due to the excellent results on correction and the real-time performance without adjusting parameters manually.