Single-shot grating-based X-ray phase contrast imaging via generative adversarial network

Talbot-Lau interferometry obtains X-ray differential phase contrast (DPC) signals of object by subtracting multiple moire patterns acquired by phase-stepping (PS) procedure. Due to the need of multiple intensity measurements, the long measuring time is inevitable in the conventional DPC imaging, giving rise to X-ray dose and fluctuations. In this paper, we propose a single-shot X-ray phase contrast imaging (XPCI) method based on deep learning. Specifically, in hardware, we propose to replace the analysis absorption grating with the high-resolution X-ray detector system to avoid the illumination flux loss caused by the analysis grating. In software, we employ a U-net based generative adversarial network (GAN) for solving the phase reconstruction problem with single shot intensity pattern. By examining the performance on a variety of simulated and experimental datasets, we demonstrate that our approach, in spite of only using single intensity pattern, is able to obtain results with high resolution and image contrast which is competitive with the conventional PS approach, while being less timeconsuming and low-dose.