A X-ray Spectrum Estimation Method by Exploring Image-domain Characteristic via CNN

X-ray energy spectrum plays an important role in computed tomography (CT) reconstruction and artifact suppression. Most methods utilize the principles of projection domain to estimate X-ray spectra, but they usually suffer from the limitations of dedicated hardware and workflow. In this paper, we propose a convolutional neural network (CNN)based method to estimate X-ray spectra from the perspective of image domain. With the potential of CNN in learning comprehensive prior knowledge from big data, the proposed method can estimate the statistical distribution of photons in each spectrum by exploring the comprehensive characteristics of CT-reconstructed images in the image domain. The proposed method estimates the spectrum by relying on the reconstructed image and requires no auxiliary hardware nor dedicated workflow. Therefore, the proposed method can be conveniently applied to spectrum-related tasks. The simulation experiments verify the efficiency of the proposed method in spectrum estimation.