Partial Convolution Network for Metal Artifact Reduction in CT Preprocessing: Preliminary Results

Laura Hellwege, Nele Blum, Thorsten Buzug, Maik Stille

Keywords: Metal artifact reduction, CT preprocessing, projection domain, convolutional neural network, partial convolution

TL;DR: A neural network with specific layer type reduces CT metal artifacts via inpainting in the projection domain.

Abstract:

Metal artifacts impair the diagnostic value of medical CT images. These artifacts occur from the projection values associated with the metal objects inside the scanned anatomy. In this work, we replace the corrupted projection values by using a deep convolutional neural network consisting of so-called partial convolution layers.
We show that the network trained on simulated data enhances newly presented projection data and therefore the corresponding reconstructed image.

Paper Type: validation/application paper

Primary Subject Area: Application: Radiology

Secondary Subject Area: Validation Study

Paper Status: original work, not submitted yet

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

Data Set Url: Data can not be provided since we are not licensed for open distribution.

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.