Spectral X-ray computed tomography image denoising using weighted local regression and noise-insensitive feature dimensionality reduction

Abstract

This study introduces a novel denoising method for spectral X-ray computed tomography (CT) images using weighted local regression (WLR). The proposed method exploits the common structural information present across different energy bins. Denoised pixel intensities of a certain energy bin are estimated using the intensities of the other energy bins via WLR. Denoising is achieved by applying a WLR model to the noisy pixel intensities of all energy bins, excluding the target bin, which obtains approximate noise-free intensities for the target energy bin. The performance of our approach was assessed using synthetic spectral X-ray CT images produced using a Monte Carlo photon simulator called the Electron Gamma Shower 5 (EGS5). Both qualitative and quantitative evaluations demonstrated that our approach effectively reduced noise across all energy bins while maintaining image sharpness. Comparisons with common denoising methods demonstrate the effectiveness of the proposed method.