3D Scout Scans Using Projection Domain Denoising

Mikhail Bortnikov, Frank Bergner, Alexey Chernyavskiy, Kevin M. Brown, Noel Black, Michael Grass

Keywords: Computer Vision, Medical Imaging, Image Denoising, Deep Learning, Computed Tomography, Low Dose CT

TL;DR: We propose a CNN based projection domain CT denoiser, which allows to do CT planning in 3D with the same X-ray dose as in 2D imaging.

Abstract: Low dose 2D scouts, also known as topograms, are commonly used for CT scan planning. Although 3D CT volumes can provide more valuable information for the selection of the scan range and parameters, the very low X-ray dose used during scout scan acquisitions results in artefacts requiring effective denoising techniques to make them useful. This has proved challenging for traditional denoising algorithms. We propose a projection domain denoiser based on a convolutional neural network (CNN), which provides improved image quality even at ultra-low dose levels. We compare two CNNs operating on two data representations, a conventional line integral data and raw photon counts, which have different quantitative properties and dynamic ranges. The results show that the two denoising strategies give rise to different properties of reconstructed images and that both projection CNNs are effective for ultra-low dose CT denoising.

Paper Type: both

Primary Subject Area: Image Acquisition and Reconstruction

Secondary Subject Area: Image Registration

Paper Status: original work, not submitted yet

Source Code Url: The source code for the experiments is an intellectual property of the company, therefore cannot be shared. In addition to this, it is not of interest without the training data.

Data Set Url: The network was trained using a private dataset, Unfortunately, due to the data usage agreement, we cannot share it at the moment.

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.