Sparse-view virtual monochromatic computed tomography reconstruction using a dictionary-learning-based algorithm

Dual-energy computed tomography (DECT) is a well-known imaging technique that can be used to differentiate and classify material composition by using projection data acquired at two different x-ray tube voltages. Dual-energy projection data can be also used to create virtual monochromatic images as the potential to reduce beam-hardening artifacts that are usually observed in conventional polychromatic images. Despite DECT’s merits, main concerns in the use of DECT in clinics may be high radiation dose imposed to patients during the examinations. In this study, we investigated sparse-view virtual monochromatic CT reconstruction using a dictionary-learning (DL)-based algorithm to provide quantitative measurements at reduced radiation dose. DL is an advanced representation learning theory that aims to find a sparse representation of the input signal in the form of a linear combination of basis elements. To validate the proposed method, we performed a systematic simulation and also we made an experiment on a skull phantom using a commercially-available dental cone-beam CT system. Two data sets of 60 projections were acquired at 80 kV p and 120 kVp separately from the system and used to create virtual monochromatic images at 90 keV and 130 keV. The image qualities were evaluated in terms of the image intensity and the peak signal-to-noise ratio.