Probe-aware self-supervised neural network for X-ray holographic reconstruction

X-ray phase contrast imaging based on propagation is a crucial technique for achieving non-destructive detection at the micro- and nano-scales. However, the recovery of phase information from intensity measurements presents a typical ill-posed inverse problem. Traditional iterative algorithms often necessitate multi-distance measurements, which increases both the complexity and time cost of experiments. Although deep learning has introduced a novel paradigm for phase retrieval, supervised methods depend on paired data that are challenging to obtain. Furthermore, existing approaches frequently ignore wavefront distortion of the illumination probe, leading to significant background artifacts in reconstruction results. To address these challenges, we propose a probe-aware self-supervised neural network (PASNet), designed to jointly estimate the complex object and probe wavefields from single-distance holograms for phase reconstruction. PASNet employs a compact U-Net architecture that takes as joint input holograms representing "object + probe" and "probe only." A physically-driven loss function is utilized to integrate the Fresnel diffraction into the network optimization process, enabling self-supervised learning without requiring labeled data. Evaluation on simulated and experimental datasets demonstrates that PASNet effectively decouples object and probe wavefronts while suppressing background artifacts. With only single-distance double-exposure, our proposed method achieves superior phase structural fidelity and background stability compared with traditional iterative algorithms and other neural networks. Additionally, robustness analysis indicates that PASNet maintains strong performance across varying propagation distances and noise levels.