Projection Data Upsampling for Sparse and Low-Dose CT Scout Scans

Frank Bergner, Rolf Bippus, Felix K. Kopp, Peter B. No¨el

Abstract:

In Computed Tomography (CT) scout scans are a first initial acquisition that determine
scan parameters and final imaging regions. Current state-of-the art scout
scans are 2D X-ray images acquired with the stationary CT scanner gantry and
the moving table. A 3D scout volume based on a low-dose, sparse-view helical
acquisition would be better for the scan planing, but image volumes usually suffer
from high noise and especially angular under-sampling artifacts. We therefore
evaluate the usage of a convolutional neuronal network to perform a non-linear
upsampling and denoising of the input projection data. We compare our approach
to a least-squares optimal linear interpolation and evaluate both algorithms with
scout scan data synthesized from real patient datasets.

Author Affiliation: Philips Research, Hamburg, Germany and Technische Universität München, Munich, Germany