Unsupervised learning for dental CT metal artifact reduction using CycleGAN with CBAM

Abstract

Metal artifact reduction(MAR) is one of the most important issues in Dental computed tomography(CT). Various methods have been suggested for metal artifact removal, among which supervised learningmethods are most popular. However matched non-metal - and metal - real CT image pairs are dicult toobtain. In this paper, we propose an unsupervised MAR method for CT using attention cycle-consistentadversarial network. The proposed method is based on unsupervised learning scheme using adversarialloss and cycle-consistency loss to overcome the none of paired data. Moreover adding the convolutionalblock attention module (CBAM) layers, we can get more improved MAR image and preserve the detailedtexture of the original image compare to standard cycle-consistent adversarial network.