Aberration modeling in deep learning for volumetric reconstruction of light-field microscopy

Abstract

Optical aberration is a crucial issue in optical microscopes, which fundamentally limits the practical imaging performance. As a commonly encountered one, spherical aberration is introduced by the refractive index mismatches between samples and environments, which will cause problems like low contrast, blurring, and distortion in imaging. Light-field microscopy (LFM) has recently emerged as a powerful tool for fast volumetric imaging. The appearance of spherical aberration in LFM will cause large changes of the point spread function (PSF) and thus greatly affects the imaging performance. Here, we propose the aberration-modeling view-channel-depth (AM-VCD) network for LFM reconstruction, which can well mitigate the influence of large spherical aberration. By quantitatively estimating the spherical aberration in advance and modeling it in the network training, the AM-VCD can obtain aberration-corrected high-speed visualization of three-dimensional (3D) processes with uniform spatial resolution and real-time reconstruction speed. Without any hardware modification, our method provides a convenient way to directly observe the 3D dynamics of samples in solution. We demonstrate the capability of AM-VCD under a large refractive index mismatch with volumetric imaging of a large-scale fishbone of largemouth bass. We further investigate the capability of AM-VCD in real-time volumetric imaging of dynamic zebrafish for tracking neutrophil migration.