Noise suppression for energy-resolved CT using similarity-based non-local filtration

Abstract

In energy-resolved CT, images are reconstructed independently at different energy levels, resulting in images with different qualities but the same structures. We propose a similarity-based non-local filtration method to extract structural information from these images for noise suppression. For each pixel, we calculate its similarity to other pixels based on CT number. The calculation is repeated on each image at different energy levels and similarity values are averaged to generate a similarity matrix. Noise suppression is achieved by multiplying the image vector by the similarity matrix. Multiple scans on a tabletop CT system are used to simulate 6-channel energy-resolved CT, with energies ranging from 75 to 125 kVp. Phantom studies show that the proposed method improves average contrast-to-noise ratio (CNR) of seven materials on the 75 kVp image by a factor of 22. Compared with averaging CT images for noise suppression, our method achieves a higher CNR and reduces the CT number error of iodine solutions from 16.5% to 3.5% and the overall image root of mean-square error (RMSE) from 3.58% to 0.93%. On the phantom with line-pair structures, our algorithm reduces noise standard deviation (STD) by a factor of 23 while maintaining 7 lp/cm spatial resolution. Additionally, anthropomorphic head phantom studies show noise STD reduction by a factor or 26 with no loss of spatial resolution. The noise suppression achieved by the similarity-based method is clinically attractive, especially for CNRs of iodine in contrast-enhanced CT.