Phase retrieval in X-ray phase contrast imaging using deep learning - Max Langer (TIMC, CNRS, UGA)

X-ray in-line phase contrast imaging is a highly sensitive imaging technique relying on the coherence of the beam to achieve contrast through interference. The development of high-flux X-ray sources has considerably advanced phase contrast imaging, pushing the attainable resolution below 40 nm in 3D and finding applications in a variety of fields. While phase-contrast imaging relies on the phase shift of the beam induced by the sample, only the intensity of the beam can be measured. Thus, the phase information is lost and must be estimated from one or several intensity images through a process called phase retrieval. Here, we consider relatively short propagation distances. Phase retrieval in this context is a nonlinear ill-posed inverse problem. Various methods have been proposed to retrieve the phase, either by linearizing the problem to obtain an analytical solution or by iterative algorithms. Recently, deep-learning techniques have yielded advances in several image processing task, specifically in inverse problems and image reconstruction. In this talk, I present our developments of deep learning-based approaches for the phase retrieval problem in in-line phase contrast, where we aim to overcome the limitations of classical approaches, such as restrictive assumptions on the forward model, the choice regularization and a priori knowledge, and the computation time.

Contact: Aurélien Gourrier