Virtual-scanning network for high-resolution light-field microscopy

Investigating sophisticated cellular and intercellular behaviors in animals is crucial to biological research, which calls for an intravital high-precision recording at ultrahigh spatiotemporal resolution. Light-Field Microscopy (LFM) achieves snapshot 3D imaging with a microlens array to uncouple the angular information, but at the cost of low spatial resolution. Recently, deep learning has revolved various microscopes including LFM with enhanced capabilities. However, deep learning-based LFM has limited performance in resolution, robustness and generalization ability. To address such challenges and expand the application boundaries of LFM-based technologies, we propose a learning-based framework, termed Virtual-scanning Network (Vs-Net) for light-field microscopy to achieve snapshot subcellular observations in vivo.