Improving visibility of X-ray phase-contrast imaging with Wiener filtering

To investigate the degrading effects of the physical parameters on the in-line X-ray phase-contrast imaging (XPCi), a simulation tool based on the Fresnel/Kirchhoff diffraction integral was firstly developed with comprehensively considering effects of the source-to-sample (S-S) and sample-to-detector (S-D) distances, the practical characteristics of a polychromatic and finite size source, the point spread function (PSF) of the fluorescent screen and the spatial resolution of the detector on the theoretical phase-contrast pattern. By a comparison between the simulative profile and the experimental one under the commonly-used parameters, an acceptable consistency has been demonstrated in despite of the deviation between the theoretically-predicted contrast (0.188) and the original experimental one (0.12). From the simulations, it is apparently observed that the fine interference pattern has been severely degraded by the finite spatial resolution, and will inevitably be further deteriorated by the system noise in practice. Since the image quality of the X-ray phase-contrast imaging is strongly dependent on the physical parameters of the system, a model-based deblurring procedure to upgrade the image visibility is preferably desired. As a simple restoration way, a Wiener filter was then introduced to offer an optimal tradeoff between the contrast preservation and the noise suppression. Finally, to minimize the deviation resulting from the finite spatial resolution, one-dimensional interpolation was performed by positioning the set square at a tiny angle to the vertical direction. The result after the Wiener-filtering-based deblurring has shown a considerably improved profile visibility: the processed experimental contrast (0.156) increased by 30% as compared to the original one (0.12) in company with the increase in the signal-to-noise ratio (SNR) by 0.9dB. With the trend of the post-filtered experimental contrast to the theoretical one, it could be motivated that higher visibility would be achieved with the introduction of more precise blurring mask and noise spectrum estimation.