Abstract

In coronary CT angiography, a series of CT images are taken at differentlevels of radiation dose during the examination. Although this reduces thetotal radiation dose, the image quality during the low-dose phases issignificantly degraded. To address this problem, here we propose a novelsemi-supervised learning technique that can remove the noises of the CT imagesobtained in the low-dose phases by learning from the CT images in the routinedose phases. Although a supervised learning approach is not possible due to thedifferences in the underlying heart structure in two phases, the images in thetwo phases are closely related so that we propose a cycle-consistentadversarial denoising network to learn the non-degenerate mapping between thelow and high dose cardiac phases. Experimental results showed that the proposedmethod effectively reduces the noise in the low-dose CT image while thepreserving detailed texture and edge information. Moreover, thanks to thecyclic consistency and identity loss, the proposed network does not create anyartificial features that are not present in the input images. Visual gradingand quality evaluation also confirm that the proposed method providessignificant improvement in diagnostic quality.

 

Introduction (beta)

None

 

Conclusion (beta)

None