Machine learning based parametric image estimation for Analyzer-based phase contrast imaging

An X-ray beam passing through biological tissue is deflected (i.e., refracted) by a small angle typically <;10 μrad. Analyzer-based phase contrast imaging (ABI) systems are capable of measuring this tinny refraction by sampling the intensity of the beam at different propagation directions. An Analyzer crystal is the key element for this task as it acts as a narrow angular filter. Since refraction effects are highly dependent of the radiation wavelength, X-ray beam must be quasi-monochromatic. Therefore the amount of photons that reach the object and detector is much lower then that in traditional radiography. Using a reasonable exposure time, noisy reconstructions of refraction images are obtained. In this manuscript, we present a machine learning parametric image estimation approach to obtain accurate refraction images from noisy raw data.