Reduction of metal artefacts in CT with Cycle-GAN

Artefacts in CT images severely disturb clinical diagnosis in medical applications. Traditional artefacts reduction methods model the physical aspects in scanning and process images accordingly. Metal artefacts are an important problem in this field. With the development of deep learning, it has bee...

Full description

Saved in:
Bibliographic Details
Main Authors: Du, Muge, Liang, Kaichao, Xing, Yuxiang
Format: Conference Proceeding
Language:English
Subjects:
Computed tomography
Deep learning
Generators
Image reconstruction
Metals
Neural networks
Training
Citations: Items that cite this one
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Artefacts in CT images severely disturb clinical diagnosis in medical applications. Traditional artefacts reduction methods model the physical aspects in scanning and process images accordingly. Metal artefacts are an important problem in this field. With the development of deep learning, it has been found that the neural networks can capture the features of artefacts. Here, we propose a Cycle-GAN method for metal artefact reduction with non-paired labels required, which much eases the work for practical problems. We validated our methods with practical CT data. From our preliminary results, our method demonstrated strong and robust ability in metal artefact reduction. Moreover, the Cycle-GAN can generate CT images with realistic artefacts at the same time, which can provide us a method for data augmentation.
ISSN:2577-0829
DOI:10.1109/NSSMIC.2018.8824544