End-to-end reconstruction of multi-scale holograms based on CUE-NET

Abstract

Digital holography is one of the most widely used label free microscopy techniques in biomedical imaging. It is an important step in hologram reconstruction to recover the missing phase information. A phase recovery method based on deep learning is presented here, which enables the phase and amplitude information of multiple holograms with different out-of-focus distances to be reconstructed quickly, and the multi-scale and large field of view can be reconstructed through the generated network model. We demonstrate the success of this deep learning multi-scale hologram self-focusing reconstruction method by self-focusing the holographic microscopic images of human red blood cells and chicken blood cells. Compared with the existing methods, this method improves the quality of the reconstructed image and the reconstruction speed. Compared with U-NET, the peak signal-to-noise ratio (PSNR) and structural similarity of the reconstructed network output are increased from 18.5 dB and 0.89 to 23.98 dB and 0.97, respectively, which can suppress the influence of speckle noise well.