Projection Domain Metal Artifact Reduction in Computed Tomography using Conditional Generative Adversarial Networks

Nele Blum, Thorsten Buzug, Maik Stille

Keywords: CT, MAR, deep learning, GAN

Abstract:

High-density objects in the field of view, still remain one of the major challenges in CT
image reconstruction. They cause artifacts in the image, which degrade the quality and
the diagnostic value of the image. Standard approaches for metal artifact reduction are
often unable to correct these artifacts sufficiently or introduce new artifacts. In this work,
a new deep learning approach for the reduction of metal artifacts in CT images is proposed
using a Generative Adversarial Network. A generator network is applied directly to the
projection data corrupted by the metal objects to learn the corrected data. In addition, a
second network, the discriminator, is used to evaluate the quality of the learned data. The
results of the trained generator network show that most of the data could be reasonably
replaced by the network, reducing the artifacts in the reconstructed image.

Paper Type: methodological development

Primary Subject Area: Image Acquisition and Reconstruction

Secondary Subject Area: Application: Radiology

Paper Status: original work, not submitted yet

Source Code Url: https://gitlab.com/maik.stille/ganmar

Data Set Url: Data is not made publicly available due to license permissions. However, data can be exchanged on personal request.

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.