KBNet-Based Noise Suppression in Edge Illumination X-ray Phase Contrast Imaging

Edge Illumination X-ray Phase Contrast Imaging (EI-XPCI) is a powerful, nondestructive imaging technique known for its enhanced sensitivity compared to conventional X-ray imaging techniques. However, EI-XPCI is prone to increased noise due to the presence of gratings that absorb X-rays, resulting in lower flux compared to conventional imaging. Noise propagates from the X-ray projections and affects the estimation of the three contrasts: phase contrast, dark field, and attenuation. Therefore, effective noise-suppression techniques are required. Despite advances in conventional X-ray image denoising methods, strategies specifically tailored for EI-XPCI remain underdeveloped. In this study, we aimed to enhance EI-XPCI image quality by adapting the Kernel Basis Network for Image Restoration (KBNet), a trained image denoising model. Our novel approach involves simultaneously inputting multiple phase step images for denoising, leveraging the inherent correlations between these images. Using our EI-XPCI simulation framework, we generated phase steps and subsequently denoised them using KBNet. The performance of KBNet was assessed in terms of the Peak Signal-to-Noise Ratio (PSNR) and the Structural Similarity Index Measure (SSIM). Our findings demonstrate that KBNet, with its multi-image input approach, outperforms state-of-the-art denoising techniques while preserving image details, offering a promising solution for noise reduction in EI-XPCI.