Abstract: X-ray phase imaging and phase tomography are popular characterization methods that can dramatically enhances signal contrast for soft materials. However, phase imaging is usually associated with spatial spectral aberration, which leads to noise amplification in the retrieval process.
In this talk, we will discuss noise-suppression techniques for X-ray phase imaging and tomography based on both the forward system design and retrieval enhancement via deep learning. For the forward system design, a structured illumination technique is used to smooth the spectral response of phase-induced diffraction and boost the low-spatial frequency signals. We demonstrate this method for a fibrous wood sphere. For reconstruction enhancement, we introduce a hierarchical architecture network that processes spectral and intensity information in a separated band. By introducing a pre-splitting to the network based on the spectral aberration, a more efficient and adaptive learning strategy is established to enable an accurate reconstruction with fewer training data and improved accuracy.
We demonstrate this method for the cellular material of a sea urchin spine. These phase imaging and tomography methods could dramatically reduce the X-ray dose and projection angles, thus enabling the in situ 4-D characterization of soft materials such as polymers, foams, and biomaterials.