Design of a light field microscope with uniform spatial resolution across an extended 3D volume

Abstract

Light field microscopy (LFM) can capture the 4D light field for high-speed volumetric imaging with single-shot image by introducing a microlens array in traditional microscopy and combing with Richardson-Lucy deconvolution algorithm, which leads to wide usage in neuroimaging and dynamics observing, etc. However, the spatial resolution of the reconstructed results shows non-uniformity along the axial optical axis due to its varying spatial-angular sampling ratio along axial axis, which limits its application in observing thick samples. In this paper, we propose a new method including light field microscope scanning strategy and a multi-image deconvolution algorithm to address this problem. First, by designing several groups of different distances between the native object plane and microlens array, we can ensure appropriate spatial-angular sampling ratio among a wide range of depth. Secondly, according to the characteristics of camera noise in the process of fluorescence microscopy imaging, we get the loss function by calculating the K-L divergence of the Poisson noise likelihood, then provide an iterative reconstruction deconvolution algorithm. Applying the multi-image deconvolution algorithm to the captured light field images, we can finally get an extended 3D volume with uniform spatial resolution. We validate our approach by using both simulation and real experiments, which contain resolution measurements of a USAF 1951 resolution target and fluorescent beads. Under the same configuration, the proposed method can reduce the variance of the resolution across 200 μm by 20%, compared with existing method, which demonstrates a more uniform resolution distribution.