Noise-Analysis-Based Non-Local Means Method for X-ray Grating-Based Mammography Denoising

X-ray grating-based mammography can potentially revolutionize the radiologic approach to breast imaging, because it works well with conventional X-ray tube and can retrieve attenuation, differential phase and dark-field information in a single scanning. However, images, especially the differential phase and dark-field images, are contaminated by noises, which would cause deterioration of the image quality and make the image denoising essential. In this paper, noise-analysis-based non-local means (NANLM) denoising method is proposed for X-ray grating-based mammography based on the previous work on noise analysis. Since pixelwise noise variances can be estimated by noise analysis, NANLM makes use of the self-similarity and sparseness in noise variances to estimate pixel intensities as a similarity-weighted average. Furthermore, in order to estimate noise variances more accurately for better image quality, a two-step NLM-NANLM method is proposed. Numerical simulation, sample model and mammography results all validate the effectiveness of the proposed method.