Dual-energy material decomposition for cone-beam computed tomography in image-guided radiotherapy

<b>Background:</b> Dual-energy (DE) diagnostic computed tomography (CT) combines two scans of different photon energy spectra which can provide additional image information as compared to standard CT. We developed a DE material decomposition scan protocol for daily cone-beam CT (CBCT) of head-and-neck patients receiving radiotherapy and tested it in a clinical trial. <b>Material and methods:</b> Our DE CBCT protocol consisted of an 80 and 140 kVp scan. The material decomposition algorithm split the low and high energy scan into components of two basis materials, aluminum and acrylic. Scans of different thicknesses and overlap of the basis materials were acquired to calibrate the model which decomposed the CBCT projections into thicknesses of aluminum and acrylic on a per-pixel basis. Pseudo monochromatic projections were created from these thicknesses and the known energy dependence of the attenuation coefficient of the basis materials. A frequency selective de-noising method was further applied to the basis material projections. The DE CBCT protocol was tested on seven patients. Two DE images were chosen, one at low (50–60) keV to evaluate soft tissue image quality and one at 150 keV to assess metal artifact reduction as compared to standard CBCT. <b>Results:</b> The de-noising algorithm reduced noise by 41% and 69% in the 60 and 150 keV images, respectively, compared to images without the de-noising. The low keV image showed an increase in soft tissue contrast-to-noise ratio of 7–43% compared to the standard clinical CBCT for six of the seven patients. The 150 keV DE CBCT image reduced metal artifacts. Enhanced streaking from metal artifacts were observed in some of the DE CBCT images. <b>Conclusion:</b> Monochromatic DE images from material decomposition can improve soft tissue contrast-to-noise ratio and metal artifact reduction. Improvements are limited, however, and new artifacts were also introduced by the DE algorithm.