Abstract
X-ray phase contrast imaging significantly improves the visualization ofstructures with weak or uniform absorption, broadening its applications acrossa wide range of scientific disciplines. Propagation-based phase contrast isparticularly suitable for time- or dose-critical in vivo/in situ/operando(tomography) experiments because it requires only a single intensitymeasurement. However, the phase information of the wave field is lost duringthe measurement and must be recovered. Conventional algebraic and iterativemethods often rely on specific approximations or boundary conditions that maynot be met by many samples or experimental setups. In addition, they requiremanual tuning of reconstruction parameters by experts, making them lessadaptable for complex or variable conditions. Here we present a self-learningapproach for solving the inverse problem of phase retrieval in the near-fieldregime of Fresnel theory using a single intensity measurement (hologram). Aphysics-informed generative adversarial network is employed to reconstruct boththe phase and absorbance of the unpropagated wave field in the sample planefrom a single hologram. Unlike most deep learning approaches for phaseretrieval, our approach does not require paired, unpaired, or simulatedtraining data. This significantly broadens the applicability of our approach,as acquiring or generating suitable training data remains a major challenge dueto the wide variability in sample types and experimental configurations. Thealgorithm demonstrates robust and consistent performance across diverse imagingconditions and sample types, delivering quantitative, high-qualityreconstructions for both simulated data and experimental datasets acquired atbeamline P05 at PETRA III (DESY, Hamburg), operated by Helmholtz-ZentrumHereon. Furthermore, it enables the simultaneous retrieval of both phase andabsorption information.