Performance of reconstruction and processing techniques for dense full-spectrum x-ray computed tomography

We have recently developed a novel x-ray computed tomography (CT) system capable of dense spectral measurements along the energy axis using a photon-counting single-pixel detector. The reconstructions have high energy axis resolution but suffer from noise due to dividing the total x-ray counts into multiple energy bins. We have recently developed two methods to suppress the noise along the energy axis: a Penalized Weighted Least Squares Algorithm (PWLS) for iterative reconstruction along the energy axis and an Eigenvector Filtering Algorithm for noise suppression in the energy domain using a set of basis functions. The two methods are briefly presented. We present a detailed comparison of the two algorithms focusing on the noise-energy resolution relationship in both methods, comparing the signal-to-noise ratio (SNR) ratio for both methods. Results show that the SNR is distributed differently across the energy spectrum for the two methods. We investigate combining the two methods forming a hybrid PWLS-Eigenvector method. The hybrid method is shown to take advantage of both methods´ SNR improvements. We conclude that both algorithms offer significant noise suppression while preserving key energy axis data.